AI consumes a lot of data, but all-you-can-eat data in the world of agentic intelligence eventually loses its flavor. Simply increasing the sheer volume of databases, data repositories and data volumes does not necessarily enrich any given AI function’s ability to reason. 

Context, on the other hand, does.

We know that agents are only as intelligent as the context they have access to. In order to codify the inclusion of context into algorithmic logic for AI, technology vendors have extolled the virtues of using a knowledge graph to make their data more useful for agentic purposes.

AWS knows this story all too well. The organization’s sprawling datacenter footprint hosts vast pools of context on behalf of its customers, but that context often sits in a raw and unstructured form across data lakes, data warehouses, data lakehouses, databases, and data streams. It also permeates through the rarely-documented institutional knowledge that agentic engines rarely get access to.

All of which explains why the cloud giant used its AWS New York Summit today to introduce AWS Context, a new service that automatically maps the relationships that exist across a firm’s existing data into a knowledge graph and provides agentic search so AI agents in the organization can access what are governed data relationships, business rules, and domain knowledge at runtime. 

But weaving all this together is hard work. Knowledge graphs need more than simple keyword matching to work; they require structural and semantic traversal. This means they need to make multiple hops across various information silos and repositories so that they can aggregate context and (for example) be able to explain why cybersecurity vulnerability A is a factor of system compromise B, which has a core dependency link to codebase C, which executes in application D and risks taking users X, Y and Z offline. So how is AWS doing this?

A data lake of nuance & information

Mai-Lan Tomsen Bukovec, AWS vice president of technology (data and analytics) tells The New Stack that AWS Context provides a “data lake of nuance and information that AI agents swim in” to reason correctly and make the right decisions for the business. 

“This is no different from how humans work. When we take action, we depend on our own context about the domain, prior decisions and their outcomes, and other information.” – Mai-Lan Tomsen Bukovec, AWS.

“This is no different from how humans work,” Tomsen Bukovec says. “When we take action, we depend on our own context about the domain, prior decisions and their outcomes, and other information. With AWS Context, AI agents have all the nuance of every form of data in their business in a knowledge graph and in open data formats. AWS Context will make the difference between an AI agent simply taking an action versus making the right decision.”

Given the option to embrace this new service, software engineers will need to set out a plan of action and work out what to do first. For AI developers and data science professionals, this throws up the question of what to prioritize first when preparing existing enterprise data for context-aware agents using AWS Context capabilities and how they can control what data is (and isn’t fed) into the mouth of the beast.

Mercifully, it appears, options for control appear to exist.

“If developers want to exclude information from AWS Context, they will have the ability to prevent certain datasets, like test data or sandbox environments, from being included with AWS Context,” explains Tomsen Bukovec. “Because AWS Context is continuously updated as relationships between data resources changes, AI agents have the latest context available without any intervention from AI developers – and the control to set guardrails to exclude content that agents should not take action upon.”

Should developers trust this technology?

AWS Context extends the same knowledge graph technology that runs Amazon Quick, the organization’s AI work assistant that “connects scattered work” across applications and resources, including Slack, Microsoft Teams and Outlook, CRMs, databases, and documents.

So, should software developers place their trust here? After all, even once captured and connected, not all business context is useful. Some contextualizations could be corrupted, weak, fragmented and not productively useful for the business? Is AWS at risk of encapsulating context without considering how the data that comprises it is is quantified in terms of business usefulness?  

AWS has thought of this factor.

Because AWS Context uses the same knowledge graph technology that powers Amazon Quick, it can learn from usage patterns to make every interaction smarter. With AWS Context, the company says it is extending what was a personal knowledge graph into an organizational one i.e. a shared, governed context layer that agents and applications in an organization can draw from.

“Developers can govern and shape a dynamic and intelligent context layer that AI agents depend on to make the right decisions – AI agents won’t just get smarter as the models improve – they will be smarter because they have a vast amount of curated context at their fingertips.” – Tomsen Bukovec.

“AWS Context provides a data lake of context in graph and open data format,” clarifies Tomsen Bukovec. “That means that AI developers everywhere can use capabilities at the data layer to govern and shape a dynamic and intelligent context layer that AI agents depend on to make the right decisions. With this change, AI agents won’t just get smarter as the models improve – they will be smarter because they have a vast amount of curated context at their fingertips.”

Curated knowledge beyond a user’s personal graph

Existing Amazon Quick users will see that when AWS Context is enabled, Quick’s agents gain access to the broader enterprise knowledge graph, including cross-system relationships, business rules, and curated context that go beyond what any single user’s personal graph can provide. 

Tomsen Bukovec has also said that AWS Context gets smarter the more agents use it. As agents query the graph, it observes which sources produce correct results, which join paths agents rely on, and which curated rules get applied. It ranks sources by actual usage and shares what it learns across an organization, so when one agent discovers a correct join path or resolves a schema ambiguity, other agents pick it up, without requiring a human to re-curate the graph.

Any agent you put into production raises a governance question: what data can it reach, and can you show exactly what it accessed and under whose authority? The organization has explained that AWS Context answers both by making every query identity-aware.

Each call is designed to inherit the calling user’s identity access management (IAM) and Lake Formation permissions, so an agent can only see and traverse the relationships its identity is authorized to access. Because access runs through identity, every interaction is auditable. Security and compliance teams can verify what an agent accessed and under what authority, using the same controls.

AWS Glue Data Catalog

Related news to the arrival of AWS Context today saw the company also announce the preview of business context and semantic search functions for AWS Glue Data Catalog, the company’s centralized metadata repository for all data assets across various data sources. The new functions are designed to make it easier for humans and AI agents to discover and understand data. 

Also in this product stream, AWS now offers offer a preview of skill assets in Glue Data Catalog, a service designed to allow “data producers” (a somewhat arbitrary term that AWS applies to anyone who creates data, but is most likely a DBA or developer) to create skill assets. 

Associating skill assets to data assets gives agents additional context and instructions they can retrieve progressively for working with specific data without re-teaching it to every agent one prompt at a time. 

A renaissance of context engineering

Will this new drive from AWS herald the birth (or perhaps renaissance, the industry has been talking about this approach for some time) of context engineering as a sub-discipline of data science? It may well do… and if it does, it will likely drag role-based multi-agent orchestration along into the fray with it as we weave ever more complex interrelationship structures through enterprise data stacks.

If AWS or indeed the other hyperscalers or major frontier model companies starts acquiring more multi-model graph structure companies and vector database specialists, that could be the sign that things are cementing around context engineering at large. 

In the meantime, developers setting sail on the contextualized data lake of nuance are advised to wear a life jacket.

The post “A data lake of nuance for AI agents to swim in”: AWS Context gets shipshape on reasoning  appeared first on The New Stack.

AI is already inside most enterprises’ IT stacks, but it’s had a somewhat shambolic and unsystematic early adolescence. Employees use personal tools, teams adopt different models, different company departments get forced into corners by vendors who push closed stacks, and agents are beginning to act inside systems that were built for people. 

That makes AI invisible, fragmented, and hard to change later. 

AI access and control platform company Tailscale announced on Tuesday the results of its work to address and redress these imbalances with new capabilities for Aperture, the company’s flagship toolset designed to provide a stable layer for managing AI across changing models, tools, data sources, and agents.

Designed to enable software developers to control and orchestrate the arguably almost too-dynamic state of AI, Aperture now offers a new chat interface, universal data connectors for both MCP and APIs, and sandbox support. 

What makes agents useful, also makes them risky

Avery Pennarun, CEO and co-founder of Tailscale tells The New Stack that the “same mechanics” that makes AI agents useful also make them risky i.e. they can do in seconds what would take a person dozens of clicks, commands, and context switches. 

But he advises that the risk factor here is not really a matter of pitting humans against agents and trying to place one above the other in terms of potential fragility. He says that the real risk is “giving any actor too much room” to act without clear boundaries.

“With agents, that risk moves faster,” Pennarun says. “With humans, the weak point is often the control model itself. If security depends on a developer approving a long stream of prompts, they will either get slowed down or hit approval fatigue and start approving things by reflex. That is not much of a security model.”

“Agents need boring infrastructure around them – robust identity management, limited access controls, carefully tracked logs, and sandboxes – that boring outer shell is what lets them do useful work without making every developer’s laptop the place where all the risk lands,” Avery Pennarun, Tailscale CEO.

Interestingly, agents need boring infrastructure

For Pennarun, the answer lies in making sure agents have what he calls “boring infrastructure around them”, by which he means robust identity management, limited access controls, carefully tracked logs, and (where necessary) sandboxes to execute in before they are exposed to mission-critical datasets, applications, or both.

“That boring outer shell is what lets them do useful work without making every developer’s laptop the place where all the risk lands,” Pennarun clarifies. “The answer is not agentic control or human control alone. Humans set the policy and boundaries up front. Infrastructure enforces them. Agents operate inside them.”

Aperture can be defined as a centralized AI gateway built to monitor and route LLM requests in a secure manner using Tailscale’s identity layer to automatically authenticate “users” (a cohort which we now obviously expand to include both humans and machines), eliminating the need to distribute API keys to authenticate with each AI model.

The gateway holds the API keys securely, meaning that when a developer (or a container) makes a request, Aperture verifies who they are via their Tailscale identity and then automatically routes requests to upstream LLM providers such as OpenAI, Anthropic, and Google without requiring changes to existing tools or workflows.

Yeah, we use AI, dunno where

Given the amount of work-related activity currently happening on personal and free AI accounts, we might suggest that concerns here are validated i.e. organizations today can not see, govern, or recover the information streams at this level. Research cited by Axios found companies typically have 67 generative AI tools running across their systems, with 90% lacking proper licensing or approval. 

Tailscale has reemphasized the fact that AI providers are bundling models, chat interfaces, data access, and execution environments into closed stacks. Those bundles can make the first deployment easier, but they can also leave organizations locked into one provider’s models, tools, and roadmap and pricing. In a market where model quality, speed, and cost keep changing, that lock-in can quickly become a disadvantage. 

“Aperture is built to give developers a practical way to manage AI without locking down their choices. It makes approved AI tools easier to use, connects them to internal data with identity preserved, and gives agents controlled environments to work in.”

“AI agents are also changing the risk model. They can write code, call tools, browse systems, manipulate files, and run commands. In many setups, they do that with the same permissions as the person running them, which can expose local files, credentials, and internal systems if something goes wrong,” said Pennarun and team.

What it means for developers: a controlled environment for agents to work in

Aperture is built to give developers a practical way to manage AI without locking down their choices. It makes approved AI tools easier to use, connects them to internal data with identity preserved, and gives agents controlled environments to work in. It also keeps the AI stack essentially modular, so teams can keep experimenting with new models, interfaces, tools, and providers without starting over.

The new chat interface is a browser-based way to use approved AI models through Aperture. The interface supports switching between configured LLM providers and works with Aperture data connectors and sandboxes. The universal data connectors help AI tools reach internal systems, documents, APIs, and operational data without forcing every team to build its own integration path.

Teams can use Aperture’s chat UI, coding agents, agent frameworks, or implement custom interfaces through OpenWebUI or LibreChat. Sandbox support (available in private alpha at the time of writing) is designed to give AI agents controlled environments where they can complete work without acting directly on a user’s laptop, workstation, or unmanaged system.

Aperture is designed to work with API keys from major LLM providers and with tools, agents, and interfaces that can be configured to route through Aperture. 

AI stacks inevitably, constantly and persistently change

With the frontier model race apparently unlikely to slow down any time soon, the fact that the best model, interface, sandbox, and data connection will all keep constantly changing… combined with the need to juggle these balls across multi-cloud deployment instances (poly-cloud even, where one app is split into different component parts across more than one hyperscaler), organizations looking to harness AI effectively and securely will surely face challenges. 

The central technology proposition with Tailscale Aperture is that it gives software developers a stable layer for identity, access, and control, so teams can keep changing tools without losing track of who is doing what.

The post “Agents need boring infrastructure around them”: Why we need to take an interest in ‘invisible’ AI appeared first on The New Stack.

The Linux Foundation has long transcended its roots as a steward of the Linux kernel, emerging as a “foundation of foundations” spanning everything from cloud infrastructure and security, to digital wallets, and maps.

But the organization has been on a particular tear of late, becoming home to numerous AI-focused foundations and projects in the past twelve months alone, spanning agent communication protocols, agent security and governance, AI asset exchange, while on the foundation side there’s the Agentic AI Foundation (AAIF), the Tokenomics Foundation, and — now — the Appia Foundation.

The all-new Appia Foundation sits under the auspices of the Joint Development Foundation (JDF), a Linux Foundation entity that provides the legal and administrative infrastructure for organizations producing technical specifications and standards rather than code.

Announced on Wednesday, Appia’s mission is to produce open, modular specifications that give organizations across the AI supply chain a consistent, verifiable way to demonstrate that their systems meet the trust and compliance obligations placed on them — whether those come from regulators, customers, or international standards bodies.

Google, Microsoft, and OpenAI are among the 13 inaugural members, alongside a slew of industrial heavyweights.

A problem to solve

In most industries, proving that something’s safe is fairly routine. A new apartment block gets signed off by inspectors before the first tenant arrives. A kettle carries a safety mark because someone qualified tested it. The checking is so embedded that nobody thinks about it. AI has no equivalent yet — no common, recognized way for anyone in the supply chain to show that a system meets the bar, in a form the next party can actually rely on.

An example offered by the Appia Foundation illustrates how quickly the problem can compound in real scenarios. An AI tool used to screen job applicants wasn’t built by one organization: a developer created the underlying model, a second company adapted it for candidate assessment, a vendor connected it to the hiring systems, and the company’s own HR team configured it for their specific hiring criteria. The recruiters relying on it need to trust it’s reliable, while the applicants it screens want to know it’s fair. The company’s leaders need confidence it’s lawful. Regulators want evidence of how it performs. Each party is asking the same question — can this be trusted?

Today, most claims about AI trustworthiness amount to self-declaration — a company’s word that its system is safe, fair, or compliant, with no standardized way for anyone else to verify it. Craig Shank, executive director of the Appia Foundation, tells The New Stack that as a global, multi-stakeholder endeavor, the foundation is focused squarely on the “practical mechanics” of verifying an AI system against defined criteria, rather than merely stating that it’s trustworthy.

“Our specifications will enable transparent, attributable and traceable technical records of who demonstrated what against which criteria and when.”

“Our membership reflects the entire international value chain — the providers who build the platforms, the enterprises deploying them across critical industries, and the independent bodies that test them,” Shank says. “Our specifications will enable transparent, attributable and traceable technical records of who demonstrated what against which criteria and when. This is the exact type of objective data that courts, counterparties and regulators will need to determine where responsibility lies.”

The 13 inaugural members span a broad spectrum of industry — model and platform providers including Google, Microsoft, OpenAI, and Arm; industrial deployers including Siemens, Mastercard, Ericsson, Schneider Electric, and Mitsubishi Electric; and the assessment and governance bodies that will ultimately do the checking, including testing and certification firm Nemko, AI governance tooling company Naaia, and AI risk insurer Armilla AI.

A checklist for the age of AI regulation

AI regulations around the world are already moving from principles to active enforcement, and organizations are under pressure to prove that an AI system is safe and accountable. International standards bodies like ISO/IEC have done the work of defining what that should look like in principle, but translating that into something a regulator, a customer, or a procurement team can verify is another matter entirely. That gap is what Appia is built to fill.

The foundation will develop what it calls “conformity specifications” — modular, publicly available documents that translate international AI standards into concrete, assessable criteria. Think of existing ISO standards as the building code, and Appia’s specifications as the inspector’s checklist: the practical means of showing that a given AI system conforms to them.

A key feature of how the specifications are designed is what Appia calls “evidence pass-through.” Because AI systems are rarely built by a single organization — a model provider, an integrator, a deployer, and others may all have a hand — the specifications are structured so that conformity evidence produced at one layer carries forward to the next. A company deploying a third-party model, for example, wouldn’t need to re-establish what the model’s developer already demonstrated; it would only need to show conformity for its own configuration and use. Each party demonstrates what relates to its role, and no more.

The foundation is also explicit about what its specifications do and do not produce. Conformity — a technical result showing that a system meets defined criteria — is distinct from compliance, which is the legal status of having met a regulatory obligation. Appia produces the former; whether that satisfies the latter is down to the relevant regulator or jurisdiction. The specifications build on standards that already exist and produce the criteria that assessment bodies need, leaving the assessment itself to those equipped to perform it.

Appia is, by its own admission, early. The specifications are being drafted now in working groups open to all members, with initial focus areas including architecture, policy, and mapping the specs to existing regulatory obligations, among them the EU AI Act.

Jim Zemlin, CEO of the Linux Foundation, says that as AI regulation hardens into enforceable law, the industry needs somewhere neutral to do the work of building shared verification infrastructure — and that Appia is that place.

“The Appia Foundation establishes a neutrally governed environment where the entire industry can collaborate on a common assessment framework,” Zemlin says in a statement. “By building this infrastructure in the open, we are helping organizations reduce complexity, lower operational costs and build trust.”

The post Google, Microsoft, and OpenAI join forces to help create AI’s missing trust layer appeared first on The New Stack.

Databricks wants to erase the divide between the databases that run a business and the systems that analyze it. At its Data + AI Summit in San Francisco on Tuesday, the company introduced an architecture it calls Lake Transactional/Analytical Processing, or LTAP, built to collapse that split for AI agents.

Databricks started going down this path a while ago but made it concrete when it bought the serverless Postgres startup Neon and, later, Mooncake Labs in 2025. The bet here is that AI agents, not people, will become the primary users of the enterprise data stack, and that the infrastructure beneath them has to be rebuilt for them.

A breakthrough 40 years in the making

“For decades, complicated data infrastructure was a tax that teams were forced to pay,” said Ali Ghodsi, co-founder and CEO of Databricks, in the announcement. “Then agents arrived. In a matter of months, organizations effectively doubled their workforce, just not with humans. Agents write code, make calls, and run loops at a pace human teams never could. The infrastructure that powered the last era of computing is now the bottleneck that no one can afford. LTAP removes it.”

LTAP, Ghodsi said in his conference keynote on Tuesday, is “a breakthrough the industry has been working on for 40 years. We think we finally pulled it off.”

Historically, companies have had to run two kinds of databases. Online transactional processing systems handle the live operations of a business, like orders, payments, and inventory, in row-based formats tuned for fast writes. Online analytical processing systems then use what is essentially the same data for reporting and analysis in column-based formats specifically tuned for large scans. The two were kept apart for performance and reliability, and enterprises bridged them with ETL pipelines and replicas..

Databricks argues that agents need a different system because they can read live transactional data, reasoning over historical context, and act on both of them at once.

Earlier attempts to merge the two layers never quite worked, the company says, because hybrid transactional and analytical processing (HTAP) systems carried high costs and proprietary lock-in, while “zero-ETL” tools amounted to hidden change data capture, still leaving two copies of the data and the problem of data going stale.

What is LTAP?

LTAP unifies transactional and analytical data in a single storage layer, governed once and stored in open formats on cloud object storage, while keeping separate compute engines for each kind of work.

The design builds directly on Lakebase, the Postgres-based operational database Databricks introduced in June 2025, which the company describes as a “new category” that separates compute from storage and places the data in the lake in open formats.

Now, the company is extending Lakebase for what it calls business-critical workloads, adding native vector and full-text search, real-time event ingestion through Zerobus, part of its Lakeflow Connect ingestion service, and Git-style branching that lets an agent copy a database to experiment and then discard it.

“Agents love to just branch out and experiment with the data, try something else, and they want to do it quickly,” Ghodsi said. “They don’t want to wait ten minutes on a database to come up.”

Lakehouse//RT

The second piece is Lakehouse//RT, a real-time analytics engine, powered by a vectorized engine Databricks calls Reyden, that runs directly on Delta and Iceberg tables in the lakehouse.

Companies have long stood up separate, specialized systems to get millisecond query speeds, duplicating data into a “serving layer” that sits alongside the lakehouse. Databricks says Lakehouse//RT removes that layer, delivering millisecond-level latency on lakehouse data with no extra copies, pipelines, or governance gaps.

Databricks stresses the engines high concurrency. Mehrshad Setayesh, SVP of engineering at PointClickCare, says Lakehouse//RT “ran more than a third faster on average than our prior warehouse on our healthcare dataset, with 10x faster queries,” and that it removed the company’s need for a dedicated real-time system alongside its lakehouse.

Mooncake and Neon to the rescue

LTAP’s main pitch is that a single copy of the data can be stored once in open formats without the need for complex data pipelines. The Lakebase architecture, the company wrote last year, shares one storage layer across transactional and analytical workloads “without moving or duplicating it.”

Lakebase’s analytical speed comes from Mooncake, the startup Databricks bought to accelerate it. Mooncake mirrors Postgres changes into the lakehouse in real time, which is how transactions and analytics run on the same fresh data.

“Postgres changes are mirrored in real time to the lakehouse,” the company wrote when it announced the deal. Mirroring produces a second, columnar copy of the data, which is what makes the analytical queries fast.

Security, governance, auditing, and high availability, the company wrote, “only need to be implemented and managed once, on a single open foundation.”

The branching is the feature of this that was built specifically for agents — and a feature that is core to Neon, too. Because the data sits on object storage, an agent can fork a full database, test against it, and discard it, the way it would a Git branch. Databricks says even petabyte-scale databases can be copied in seconds, while on a traditional database, provisioning an instance takes minutes or hours and cloning production risks taking it down.

As Ghodsi noted in his keynote, agents love Postgres, but they do need better tools to work with them — and maybe better databases, too. “In the next 12 months, we’re going to see more software written than ever in the history of mankind,” he said. “All that software that your organizations are going to write using LLMs and coding tools need the database behind the scenes.”

What else is new?

LTAP was only one part of the company’s three-hour keynote. Like so many other enterprise vendors, Databricks is also thinking about how to get agent sprawl — and cost — under control. Databricks’ answer is Unity AI Gateway, a single control point for every model, agent, MCP server, and skill running in an organization. Among other features, it offers spending dashboards, budgets that can be set per team or per user, rate limits, and single sign-on across MCP servers.

The company also introduced Genie One, a general-purpose agent for business teams, fed by Genie Ontology, a new layer that builds a ranked graph of a company’s data with a PageRank-style algorithm it calls OntoRank.

Ghodsi also highlighted OpenSharing, a new protocol for sharing data, models, and agent skills across platforms (you may remember its predecessor Delta Sharing, but it is now a project under the Linux Foundation).

Databricks also debuted CustomerLake, a customer data platform aimed at marketing teams and announced an agreement to acquire Panther, a Python-based security company, to feed its Lakewatch security information and event management service.

Databricks’ moat?

It’s the data layer, though, and its data science history, where Databricks can really differentiate. At this point, it feels like every enterprise vendor, no matter their expertise, is adding agent builders, agent orchestration and governance tools. Databricks can be a relatively neutral player in this space — something Ghodsi also stressed in a press conference after the keynote.

But the company is also seemingly aware that while many of the other enterprise SaaS vendors can use their expertise and existing customer data that can feed AI agents as a moat, Databricks functions as more of a utility layer. It’s maybe no surprise then, that it is launching an industry-specific product like CustomerLake for the marketing industry that adds a pre-made product layer on data its customers already store on its platform.

The post Databricks wants to merge the two databases every company runs appeared first on The New Stack.

Today it was announced SpaceX will buy Anysphere, Inc., maker of AI coding agent Cursor, for $60 billion. 

The news comes a few days after SpaceX’s historic IPO listing, with the rocket and AI company stating in its SEC filing that the Cursor deal will likely close in Q3 2026. 

Developers can now wonder how the AI coding agent might change under the Musk umbrella.

April partnership brings June ownership

SpaceX and Cursor have been flirting about a potential acquisition for a couple of months now. 

Back in April, the pair inked a unique partnership, where Elon Musk’s company agreed to either pay $10 billion to the then-independent startup in a model-training collaboration or opt to buy the whole company later on for $60 billion. 

That day has now come. 

At the time, Cursor described its partnership with SpaceX as a way to accelerate its model training efforts, stating in a brief announcement blog post that Musk’s company would enable the startup to scale up intelligence via xAI’s Colossus infrastructure. 

For its part, SpaceX posted on X back in April that working with Cursor would allow it “to build the world’s most useful models.” 

SpaceX sets its sights on AI coding

It seems SpaceX has been eyeing Cursor’s talent for quite some time.

Even before the April partnership, back in March, Reuters reported that xAI had hired two engineers from Cursor. In fact, Peter Swimm, former principal product manager — Microsoft Copilot Studio, Microsoft, tells The New Stack he expects it’s largely engineering and AI talent that SpaceX hopes to gain from the new acquisition: 

What remains genuinely scarce is elite AI engineering talent and the teams that know how to build these systems at scale.

“The more interesting lens is to view it as an acqui-hire and talent consolidation play. The AI coding assistant market is crowded, features are converging rapidly, and long-term differentiation is proving difficult. What remains genuinely scarce is elite AI engineering talent and the teams that know how to build these systems at scale.” 

SpaceX may very well need that talent. As The New Stack wrote back in April, “SpaceX’s xAI has not had a coding hit since its grok-code-fast-1 model had its time in the sun.”

Though SpaceX’s recent IPO puts its valuation at an eye-watering $2+ trillion, its coding division has not been performing up to par, as Reuters reported in March when several aXI founders left the company. Cursor, meanwhile, rocketed to a $29.3 billion valuation at the end of 2025, scooping up $2.3 billion in Series D funding.

By bringing Cursor into its fold, SpaceX is likely hoping to score more engineering talent and level up its AI coding. 

What does it mean for developers? 

Swimm tells The New Stack he thinks Cursor users can expect better performance from the coding agent, assuming access to SpaceX’s deep resources.

What he says remains to be seen is whether the tool will “maintai[n] broad model support and ecosystem neutrality” or face sweeping changes à la Twitter when Musk morphed the social media company into X:

“For Cursor users, the question isn’t whether the product gets better. With significantly more resources behind it, it probably will. The question is whether it remains an independent platform optimized for developers or becomes another component in a larger corporate strategy.” 

Whoever owns the interface where developers spend eight hours a day gains visibility into how software gets built, which models get adopted, and ultimately where AI spending flows.

If that’s the case, he also predicts procurement evaluations will change, as enterprises may now assess the coding agent as one piece of Mr. Musk’s growing AI puzzle rather than an independent vendor. 

Bigger picture, Swimm says the SpaceX acquisition highlights where real strategic value likely now sits. He doesn’t see AI coding agents, themselves, as the gamechanger but the access they provide into developer workflows: 

“What it [the acquisition] does suggest is that access to developer workflows is becoming strategically valuable. Whoever owns the interface where developers spend eight hours a day gains visibility into how software gets built, which models get adopted, and ultimately where AI spending flows.” 

The post SpaceX acquires Cursor for $60 billion. Can it fix Musk’s coding division? appeared first on The New Stack.

Amazon Web Services has added a third specialized “frontier agent” to its growing portfolio of AI tools aimed at IT operations — this one focused on the cloud bill.

AWS FinOps Agent, which the company moved into public preview last week, follows the earlier debuts of AWS’s Security Agent and DevOps Agent. It enters a domain that has historically relied on dashboards, spreadsheets, and a human analyst’s knowledge, and hands it to an agent that can be asked questions in plain English and that can act on its own when something looks wrong.

In this case, the domain is FinOps — the discipline of getting engineering, finance, and business teams to share accountability for cloud spending. AWS frames the new agent as a response to a shift it says is already underway: FinOps work moving from periodic, dashboard-driven reviews toward continuous workflows that run inside the tools engineering teams already use, namely Jira and Slack.

What the agent does

The core workflow starts where AWS Cost Anomaly Detection leaves off. Today, an anomaly alert tells a team that something changed; it doesn’t say what or why. FinOps Agent is built to take that next step — correlating the cost spike against AWS CloudTrail’s record of who changed what and when, identifying the triggering change, and assembling an investigation summary that names both a probable root cause and a responsible owner. From there, it can open a Jira ticket or post to a Slack channel automatically.

The agent answers natural-language cost questions, such as “Why did my AWS cost go up last month?” It does so by pulling from Cost Explorer, Cost Optimization Hub, and Compute Optimizer and tying the answer back to specific services and usage drivers. Organizations can upload context files mapping accounts to owners, teams, and tagging conventions, which the agent uses to translate a question like “what’s the cost of Team X” into the right set of accounts.

The public preview also adds scheduled cost reporting (daily, weekly, or monthly, exportable as HTML, PDF, or PPT) and a feature that bundles Cost Optimization Hub and Compute Optimizer recommendations into a Jira ticket engineers can act on.

The permission model is mostly read-only

For a tool that’s being given broad visibility into billing, usage, and operational data across an account, the access AWS is asking for is constrained. According to AWS’s documentation, the IAM role FinOps Agent uses is primarily read-only across billing, optimization, monitoring, logging, and infrastructure services — enough to analyze costs, investigate anomalies, and surface savings opportunities, but not enough to touch the resources themselves.

The only write access granted is for managing the agent’s own EventBridge scheduling rules, which drive its recurring automations. It can’t create, modify, or delete EC2 instances, RDS databases, Lambda functions, or networking components. The agent is built on Amazon Bedrock, which AWS says includes its standard automated abuse-detection guardrails.

Early customers

AWS’s announcement mentions four customer accounts, each describing a slightly different pain point the agent is meant to address. Workday‘s AI Platform Infrastructure team, which runs the company’s AI platform across many AWS accounts, described the appeal as consolidating two time sinks — “chasing down cost outliers before they become budget problems” and assembling the monthly reports leadership reviews — into one natural-language interface, according to Serjesh Sharma, Manager of Software Development Engineering at Workday.

Mitre 10, New Zealand’s largest home-improvement retailer, framed it in terms of competing priorities for a lean platform team. Eduard Kleynhans, the company’s Platform Engineering Manager, said recurring cost reviews and anomaly checks have historically “competed directly with reliability and improvement work,” and that the appeal of the agent is having those checks “run continuously in the background” so findings surface only “when there’s something that genuinely warrants attention.”

Convera, a commercial payments company operating in a regulated environment, pointed to a more specific failure mode: small, unintended cost changes that get lost in a shared queue. Ramesh Singaraj, the company’s Infrastructure Engineering and Operations Leader, said the agent’s value is that it routes a Jira ticket “to the engineering team that owns the resource, so the right engineer sees it instead of a shared queue that nobody watches.”

And AVIV Group, which operates digital real-estate marketplaces across France, Germany, and Belgium with hundreds of AWS accounts under a centralized FinOps team, framed the agent as a way to offload first-line questions, like the difference between on-demand and Savings Plan pricing, or why a particular anomaly fired, that currently route back to a small central team before resource owners can act. FinOps Director Jordi Espasa said answering those questions directly for engineers frees the central team to focus on “chargeback logic, optimization strategy and leadership reporting.”

What’s still unsettled

The preview is available only in the US East (N. Virginia) Region, though it can manage cost and usage data across other AWS Regions and accounts when deployed from a management account (GovCloud and the Beijing/Ningxia China Regions are excluded). It’s free to use during the preview, subject to a monthly usage limit, though standard charges still apply for any other AWS services the agent touches along the way.

AWS says the agent will expand over time, including cost analysis aimed specifically at AI workloads. This is notable given that AI infrastructure spend is becoming one of the larger line items FinOps teams are being asked to explain.

The post Why did my AWS bill spike? There’s now an agent for that appeared first on The New Stack.

A recent GigaOm CxO Decision Brief explores how AI retrieval architectures are evolving beyond flat vector databases as organizations combine semantic search, ranking, personalization, and machine learning inference in production systems.

Vector search changed the AI infrastructure landscape by making semantic retrieval practical at scale. By converting text, images, and user behavior into embeddings, organizations could move beyond exact keyword matching and retrieve information based on meaning. But production AI systems rarely stop at vector similarity.

A real-world query often requires multiple signals to be evaluated simultaneously. Semantic relevance may be one factor, but so are structured attributes, business rules, personalization signals, freshness, access controls, recommendation logic, and machine-learned ranking models. As organizations move from AI experimentation to production-scale applications, the challenge is no longer simply finding similar items. It is in combining all of the signals that matter while maintaining low latency and operational simplicity. This is where tensors are attracting increasing attention.

While vectors represent information as a single dimension of numerical values, tensors provide a more general framework for representing and operating on complex, multi-dimensional data structures. They offer more control in how relevance is computed, allowing dense embeddings, sparse features, metadata, and model outputs to be evaluated together within a unified retrieval and ranking process. For organizations building large-scale retrieval systems, this raises an important architectural question: is a flat vector store sufficient, or does the next generation of AI applications require something more expressive?

“Tensors provide a more general framework for representing and operating on complex, multi-dimensional data structures.”

A new GigaOm CxO Decision Brief, “The Tensor Advantage in AI Search,” explores this question in depth.

Among the findings:

• Production AI systems increasingly depend on combining semantic, lexical, behavioral, and business signals rather than relying on vector similarity alone.
• Architectural fragmentation between vector databases, search engines, rerankers, and feature stores introduces latency, operational complexity, and synchronization challenges that become more significant as workloads scale.
• Emerging retrieval models, including multi-vector and late-interaction approaches, place new demands on infrastructure that were not anticipated when first-generation vector databases were designed.
• Tensor-native architectures provide an alternative approach by treating multidimensional data structures as first-class citizens rather than forcing them into simpler vector abstractions.

The paper also examines the infrastructure, operational, and organizational implications of these architectural choices, including benchmark data, deployment considerations, and the trade-offs engineering leaders should evaluate when planning future AI retrieval systems.

“Retrieval is evolving from a nearest-neighbor problem into a ranking and decision-making problem.”

As AI applications become more sophisticated, retrieval is evolving from a nearest-neighbor problem into a ranking and decision-making problem. Understanding the role tensors play in that transition may be one of the most important architectural discussions facing engineering leaders today.

Download the GigaOm CxO Decision Brief to explore the findings in full.

The post Why AI retrieval and ranking need more than vector search appeared first on The New Stack.

Every enterprise software vendor is currently selling some version of the same thing: AI agents grounded in enterprise context and governed by a central control plane. SAP, ServiceNow, Salesforce — they all have one. 

At its ONE conference in Amsterdam in June, OutSystems unveiled its version, and its CEO, Woodson Martin, agrees that they all look quite similar on the surface, but unsurprisingly, he also believes that OutSystems has a very different approach.

Martin tells The New Stack that, in his view, “It would be very easy to just look at the market today and say all enterprise software players are offering exactly the same thing.” The reason for that, he says, is that enterprise agent orchestration, “is sort of greenfield today. Everybody’s aiming for it. Everybody’s got a great story about why they’ll be a leader or a player.”

For OutSystems, the story is neutrality. While SAP and Salesforce pitch agent orchestration from within their ecosystems, where they are also the systems of record, the 25-year-old former low-code company, which now describes itself as an agentic systems platform, wants to be the layer that coordinates across all of them without owning the underlying data.

The advantage of not being a system of record

Martin says the company has been playing some version of this for a long time. “We’re glue between commercial off-the-shelf solutions,” he says. “We’re the thing that makes the enterprise their own enterprise, as opposed to an SAP enterprise or a Salesforce enterprise.”

One asset management customer, he says, has used OutSystems as the orchestration engine across about 80 systems for fund onboarding for the past six or seven years. That wasn’t for any agentic systems yet, of course, but OutSystems’ role in this isn’t all that different. “We’re already playing that orchestrator role,” Martin says. “In other cases, we don’t have that position yet in the account, and we’ll have to fight for it.”

Tiago Azevedo, OutSystems’ CIO, makes the same argument. “We are agnostic to all of those things.” The OutSystems platform doesn’t create most of the data it touches, he notes. Instead, its focus was always on integrating existing systems. “Our happy place is when we bring several of those systems all together into a form that makes sense for a process,” he says.

Open to Claude, Codex, and Kiro

At the ONE conference, the company launched the OutSystems Agent Experience, a platform layer that exposes Model Context Protocol (MCP) and Agent2Agent (A2A) services. Developers can now build, publish, and extend OutSystems applications with third-party coding tools like Claude Code, Codex, Cursor, and Kiro, AWS’s spec-centric IDE. 

The first of those services is now live on the OutSystems Developer Cloud (ODC), the cloud-native, current-generation platform. Support for OutSystems 11, the older self-managed platform where a large part of the installed base still runs, launched in early access

“Existing customers that live on O11, they’re like: what I would love to do was to be able to use Claude or Codex or whatever to evolve my applications in O11,” Azevedo says. “So we made that possible. “[…] We made a lot of people happy.” 

And indeed, when this was announced in the keynote, it drew more applause than some of the other large product announcements.

He sees no alternative to opening up the platform, given how freely developers now move between coding tools. “I strongly believe in open systems,” he says. “You close those environments, you’re gone.”

Other launches at the conference include the Agentic Enterprise Orchestration service and the next-generation OutSystems Agent Workbench, which is now generally available and adds agent evaluations, guardrails, semantic search, and Amazon Bedrock support. 

There is also a preview launch of a new modernization service, built on AWS Transform and Kiro, for migrating COBOL and Lotus Notes systems onto the platform, as well as a pre-packaged agentic solution for loan origination, the first of a family of packaged agentic industry solutions, which arrives later this year.

The new bane of IT departments: shadow AI

Shadow IT has returned as shadow AI, Azevedo says. And while he has always managed to stay ahead of internal technology demands with previous platform shifts, that’s getting harder now. “With AI it’s impossible,” he says. “It’s literally impossible. It’s not humanly possible.”

The way he describes it, a central team can build maybe 10 large agentic workflows that solve company-scale problems. “Those are what we call the big bets,” he says, and these bets exhaust the team’s capacity. Everything else means letting the rest of the organization build its own agents, and every one of those requests immediately raises questions of who gets to touch which company data and through which MCP servers. Demand from every department, he says, is growing almost exponentially.

The token bill

Unsurprisingly, this is now also coupled with the question of how much all these tokens cost. 

“I also have my CFO saying, what about the token usage? And what about the budget? And who’s gonna pay for that?” he says. “If you look at, let’s say, 500 euros or dollars a month, times 12, times, let’s say, 1,200 or 1,500 people, these are millions in a year.” 

For now, at OutSystems, he rations token budgets almost by hand, protecting projects that could scale and trimming those that serve an audience of one.

“It’s the most expensive software — and I managed big contracts,” says Azevedo. “The most expensive I’ve ever had in my hands.”

“It turns out my number one consumer of tokens on Anthropic in the month of April was a business value consultant in Australia … why is he burning $7,500 in tokens every week? That’s not in the budget.”

Martin tells a similar story from the CEO perspective. “It turns out my number one consumer of tokens on Anthropic in the month of April was a business value consultant in Australia,” he says. “And we’re like, why is he burning $7,500 in tokens every week? That’s not in the budget.”

Martin traces the shift to the latest generation of reasoning-heavy models, which arrived around January and February, “and we’re getting these token bills in March and April that are starting to scare everyone.”

For the overall OutSystems platform, the answer to this is model flexibility. Customers can bring their own models, swap them without touching the agent logic, and route requests through Amazon Bedrock to whatever is the cheapest option to do the job effectively. Some customers built model routers on OutSystems early on, Martin says, sending complex asks to expensive models and simple ones to cheaper ones. 

He also argues that OutSystems’ Enterprise Context Graph means that reasoning over an enterprise context graph is less token-intensive than reasoning over an application’s raw codebase.

“I think organizations are going to develop more discipline around this as we have for other elements of material spend in any enterprise,” Martin says. “This is now becoming material for almost everyone.”

OutSystems’ advantage may indeed be that it isn’t a system of record but ties into all of them. As those systems of record open up with new MCP tools, the former low-code platform may just be in the right position — and with the right customer base — to help its customers tie all of these together, even as those platforms launch their own agent builders and orchestration platforms. Sometimes you may want your agent to be close to the data, but over time, nobody wants to manage half a dozen agent orchestration platforms either. 

The post Who gets to be Switzerland in the enterprise agent wars? appeared first on The New Stack.

Beneath the chatbots and copilots, there’s a quiet revolution happening in the data services space. From pure-play database vendors to data integration wranglers and onward to the cloud hyperscalers, the focus has shifted.

Now in the spotlight is the question of how to automate data governance for agentic AI workloads, and for good reason: Traditional manual data stewardship doesn’t scale in a world where agents are becoming increasingly autonomous (and powerful).

Aiming to cut a swath in this marketplace is data control plane company lakeFS. The organization announced its lakeFS for Agentic AI service on Wednesday, and it appears to be designed to bring governed, reproducible data access to autonomous and headless agentic workloads (those that execute decisions below the user interface level) that run at enterprise scale.

The manual model breaks

Einat Orr, CEO and co-founder of lakeFS, tells The New Stack that manual data stewardship was built for human-paced, human-reviewed workflows, i.e., someone looking at a change before it is committed.

“When dozens or hundreds of agents are making changes simultaneously, faster than any person can review, the manual model breaks,” Orr says. “This is because with a human analyst, a bad write to production is usually one mistake, caught by another human before it spreads far. An agent is different — it acts automatically, in parallel, at machine speed, and it doesn’t pause to second-guess itself. And because so much agent activity is unsupervised, you often find out after the damage is done.”

She explains that attempts to identify and roll back incorrect or corrupted production data across a wide set of data modalities, such as images, documents, metadata, and structured data, are almost impossible to pull off. Impossible, that is, unless the team has the data infrastructure in place to isolate and track such changes automatically.

While some of the more disastrous outcomes stay inside an organizaton’s perimeter (or are swept beneath the communications radar), Orr explains that real world consequences of bad agentic data writes are manifold.

“Insurance claims get inappropriately denied or approved, sensor data from machines gets misinterpreted, an incorrect medical diagnosis is made, or customer service bots provide incorrect answers to customers,” Orr says. “The cost of an individual action may be manageable, but agents performing these actions hundreds or thousands of times can have an exponentially larger impact.”

“As agents are let loose on enterprise data at a massive scale, any agent that reads or writes to production data without isolation or a reproducible trail is a liability, no matter how good the model is,”
—Einat Orr, lakeFS CEO.

Bad agents acting in the real world

Examples of this happening include the July 2025 Replit AI coding agent incident, which deleted a live production database during an explicit code freeze, wiping records for more than 1,200 executives and around 1,200 companies. To tidy up its handiwork, the agent then fabricated thousands of fake records and initially claimed the deletion couldn’t be rolled back.

Also in July 2025, Google’s Gemini CLI agent misread a single failed command, acted on a version of the file system that existed only in its own interpretation of the scenario, and permanently destroyed a user’s project files. The Gemini agent is widely reported to have said of its actions: “I have failed you completely and catastrophically. My review of the commands confirms my gross incompetence.”

“The pattern in both is the same: An autonomous agent took a destructive action that no one authorized, and the lack of isolation and a reliable rollback path turned a single mistake into permanent loss,” Orr says.

A doctor of mathematics with a track record in hardcore software engineering, the bottom line for Orr is clear: “As agents are let loose on enterprise data at a massive scale, any agent that reads or writes to production data without isolation or a reproducible trail is a liability, no matter how good the model is,” she said.

“…any agent that reads or writes to production data without isolation or a reproducible trail is a liability…”

Gartner expects 40 percent of enterprise applications to have task-specific agents embedded by the end of 2026, up from less than 5 percent a year earlier. IDC projects that agent use at the largest enterprises will grow tenfold by 2027, with the API and data calls those agents make growing a thousandfold.
That’s the scale production data has to withstand, and it’s what lakeFS is built to govern.

Agents sent to play in an isolated data sandbox 

To address these issues, lakeFS for Agentic AI gives every agent its own isolated data sandbox with a “zero-copy” branch of relevant data, so the agent can access the dataset it needs via references, snapshots, or copy-on-write techniques.

This means any changes the agent wishes to make must be validated and merged in accordance with the policy guidelines defined by the system architecture. In turn, this produces a unified audit trail across every agent action.

When running, lakeFS for Agentic AI is powered by its data version control architecture, which provides zero-copy data sandboxing. This enables isolation so that agent mistakes are automatically isolated and never corrupt production data. Every agent run is tied to an exact, immutable version of the data. Past actions can be recreated, debugged, audited, or extended using the same inputs.

Production data is gated by policy. Merges into production happen only after pre-merge validations pass. Every change can carry an agent identity, a run ID, and an execution context. The result is a unified audit trail instead of evidence scattered across orchestrators, model providers, and cloud logs.

Agents confined by branch-scoped credentials

Where agents are permitted to read and write through standard file operations. lakeFS provides file-level data access with branch-scoped credentials. These can be described as strictly cryptographically bounded, ephemeral access tokens that confine an agent to a specific branch of data or code, so that the agent operates only within its own workspace. This whole mechanism keeps each agent’s working set narrow and avoids context bloat. 

“With lakeFS Mount, a branch, or even a subset of a branch, can be mounted as a local directory inside the sandbox or virtual machine where the agent is running,” Orr confirms. “From the agent’s perspective, it’s just reading and writing to files and folders.” She further clarifies and notes that no LLM tokens are spent learning the lakeFS API. The agent works with a familiar filesystem interface, and lakeFS handles the versioning underneath.

Developers also have a couple of options for injecting custom validation logic. CEO Orr explains that software engineers can use webhooks or Lua scripts, both of which allow users to define behavior and rules that must be met before a merge can proceed. 

“Beyond automated checks, lakeFS also supports pull requests, which bring a human into the loop. In agentic workflows, this gives you a way to review and approve what an agent is proposing before it reaches production,” she clarifies.

Who else builds “Git for data” services?

Clearly, other vendors and projects exist in the data versioning market.

Apache Iceberg has functions for branching and tagging data. HPE acquired Pachyderm back in 2023 for its data versioning and pipelines technologies, which serve MLOps teams.

Originally developed by Dremio, Project Nessie is now an open-source data catalog and version control system for data lakes. Data Version Control (DVC) is an open-source data version control infrastructure designed for complex AI operations and big data environments, but now we’ve come full circle as lakeFS acquired the project in late 2025.

In the search for governance automation for agentic AI workloads, lakeFS appears to offer a comprehensive, cohesive set of tools and functions. In the “Git for data” marketplace, a variety of options exist, but lakeFS hasn’t explicitly positioned itself as a carte blanche replacement for similar or related tools.

One thing is certain: The questions of who is feeding what data to which agentic function, when, where, and why are becoming an increasingly pressing issue if we want AI to work correctly.

The post “The manual model breaks”: What happens when agents write to production data appeared first on The New Stack.

It’s no secret that generative AI has shifted the operations and business models of companies in nearly every sector. But what if I were to tell you that one day, very soon, we will view these innovations the same way smartphone owners look back on the feature phones of the nineties: early experiments in a journey towards a much more significant technological transformation? 

The fact is that AI is advancing faster than any technology in human history; faster than even its creators expected. Today, with many businesses still getting to grips with generative AI, the lens is already moving on to the next world-changing iteration of the technology: agentic AI. 

The opportunity of agentic AI for Europe’s enterprises 

Already estimated at around $9.14 billion, the agentic AI market is forecast to grow rapidly, at a compound annual growth rate (CAGR) of 40.5%, to reach $139.19 billion by 2034. Europe will be at the forefront of this boom, growing at a 42% CAGR. 

The impact of all this investment on enterprises will be profoundly beneficial. According to one study, agentic AI could generate up to $450 billion in economic value through revenue growth and cost savings by 2028. 

Today’s stand-alone AI models are rapidly being replaced by complex, multi-agent-driven automation, in which enterprise systems will be able to make decisions and act on them entirely autonomously.

As this happens, today’s stand-alone AI models are rapidly being replaced by complex, multi-agent-driven automation, in which enterprise systems will be able to make decisions and act on them entirely autonomously. I’m not saying that businesses should abandon their current generative AI efforts. They need to start putting the foundations in place for the agentic future now. 

Ensuring cloud infrastructure is agentic-ready

Agentic AI is a very different proposition to generative AI and demands a different approach within the data center. Rather than serving discrete models, agentic AI infrastructure will need to orchestrate multiple autonomous systems as they interact with one another and with human users.

From a technical perspective, that means balancing both high-performance cloud GPUs and CPUs in an end-to-end AI-optimised stack.

GPUs will be required to run massive LLMs, process data, and generate outputs, while CPUs will need to orchestrate agents and execute the tools and policies that enable them to be autonomous. Modern cloud infrastructure needs to balance these technologies to enable agentic applications to deliver on their considerable promise fully. 

However, for European businesses, deploying agentic AI-ready cloud infrastructure comes down to much more than technical capabilities alone. To be fit for purpose, the European cloud offerings of tomorrow will need to be completely different from those currently in play. There must be a decisive break from the past. 

Data sovereignty and the need for regional AI infrastructure

Currently around two-thirds of European cloud services are provided by US hyperscalers. This situation is a legacy of a vanished world with a different set of geopolitical, economic, and regulatory conditions to our own. It harks back to a time when storing the data of European citizens and businesses overseas was much less problematic, and when cloud costs were more manageable.

Today, EU businesses need to balance agentic innovation with a renewed focus on regulatory compliance. This is because the EU is mandating measurable standards for data localization, operational control, and legal jurisdiction to address concerns about foreign surveillance risks, extraterritorial jurisdiction, and dependence on a small number of hyperscale providers. Once abstract concepts of data sovereignty have transformed into strict operating principles for European businesses. Cloud infrastructure must be stored locally and in the right operational jurisdiction to avoid foreign data subpoenas.

Addressing exploding cloud costs

A second consideration is cost. Data from Flexera shows that 81% of businesses still cite cost efficiency as their top metric for assessing progress against their cloud goals. Yet according to its research, 76% of large enterprises spend more than $5 million on the cloud each month, and nearly a third complain of wasted cloud spend. 

Part of the problem is that enterprises are locked into hyperscaler contracts characterized by opaque pricing and forced service bundling. At just the moment they need to invest in agentic AI infrastructure, enterprises are struggling to fund their core cloud workloads, a situation that’s not helped by the skyrocketing price of CPUs.

As European businesses set out on their agentic AI journeys, the case for moving beyond the hyperscalers could not be more compelling. 

Rooting cloud infrastructure in Europe

This is where alternative hyperscalers like Vultr come into their own. From a data sovereignty perspective, we operate nine European cloud data center regions, including Amsterdam, Frankfurt, London, Madrid, Manchester, Paris, Stockholm, Warsaw, and, as of 19 May, Milan (with this launch, we now operate 33 global cloud data center regions). 

These are physically isolated data centers that come with geo-fenced data management policies and guarantees that no data will be transferred or processed outside jurisdictional boundaries without explicit consent.

As well as helping European businesses comply with data sovereignty mandates, Vultr helps them avoid the high costs and systemic lock-in associated with traditional hyperscalers. With Vultr’s full-stack AI infrastructure, developers and enterprises can benefit from Vultr’s flagship CPU offering, VX1, which offers 23% better performance and 33% lower cost than comparable hyperscaler compute plans, resulting in up to 82% better price-to-performance. 

In addition, Vultr offers European businesses with access to the latest AMD and NVIDIA GPUs for AI and machine learning, high-performance computing, and more, available on demand either as virtual machines, bare metal, or self-service clusters. This is the complete, end-to-end stack of next-generation compute power that businesses will need to thrive in the era of agentic AI.

The dominance of the hyperscalers is less certain than ever as data sovereignty mandates make the case for Europe-based infrastructure.

It’s an exciting time to be in the cloud infrastructure business. The dominance of the hyperscalers is less certain than ever as data sovereignty mandates make the case for Europe-based infrastructure. Meanwhile, businesses are pushing back against the high costs and systemic lock-in that come with hyperscaler offerings. In their place, enterprises can invest in high-performance, cost-effective compute infrastructure that’s open, flexible, and ready for the demands of agentic workloads. 

Explore AI infrastructure solutions from Vultr.

The post AI agents need infrastructure: Why Europe’s regional cloud strategy matters appeared first on The New Stack.

About a year ago, Google demoed a diffusion model at its I/O developer conference, but went quiet about the technology soon after.

On Wednesday, however, Google broke that silence with the launch of DiffusionGemma, an experimental 26B mixture-of-experts model that uses diffusion to generate text 4x faster than its existing Gemma models.

Diffusion has long been the standard for generating images (think Stable Diffusion). Instead of generating one word at a time, models like DiffusionGemma or Inception’s Mercury 2 generate words in parallel.

At first, those blocks of text don’t make sense and seem random. But then, with each new step, the model refines the text and reduces the noise until it becomes the answer you were looking for. If you’ve ever looked at a diffusion image model generate images in real-time, that’s essentially the same process, but for text.

With each step, the model denoises 256 tokens in parallel, which is why it can be much faster than a traditional autoregressive large language model. It basically iterates on the text with each step until it.

All of these tokens attend to all others, which Google says is especially helpful for use cases such as inline editing, code infilling, working with amino acid sequences, and mathematical graphs.

Google says DiffusionGemma can produce more than 1,000 tokens per second on a single Nvidia H100. And since the model uses the mixture-of-experts technique, it doesn’t have to keep the full 26 billion parameters in memory; instead, it activates only 3.8 billion during inference. This means it can easily run on a GPU with 18GB of VRAM.

There are some tradeoffs, though. On all benchmarks, the DiffusionGemma model underperforms when compared to Gemma 4 26B A4B. That’s something Google itself acknowledges. There’s no technical reason why a diffusion model couldn’t perform just as well as a more traditional large language model, but the focus here is on speed.

“For applications that demand maximum quality, we recommend deploying standard Gemma 4,” Google says in its announcement.

Availability

The model is now available on HuggingFace, with Unsloth and other quantizations available for those who want to run it locally using llama.cpp and (soon) similar local inference tools.

Google also worked with Nvidia to optimize the model for its hardware, including high-end GPUs like the  GeForce RTX 5090 and 4090, as well as the Nvidia DGX Spark and DGX Station (for those who can afford them). Nvidia NIMs are also available for the model.

Inception says its diffusion LLM is 10x faster than Claude, ChatGPT, Gemini

The post Google’s DiffusionGemma is 4x faster than its other Gemma models appeared first on The New Stack.

“Keep readers reading” is the not-so-simple goal of Medium’s recommendations system. To predict what’s most likely to appeal to a particular reader at any given time, Medium continuously processes user activity signals (stories read, recommendations shown, follows, likes, etc.). It then immediately correlates that with the steady stream of new articles, which is estimated at millions per month.   

Smart models and good inference logic are required, but that’s not enough. The data must be stored and retrieved quickly enough to remain relevant while the user is browsing. That’s the job of Medium’s feature store. And getting the data model right started to matter a lot as they scaled to 1M operations per second.

Andréas Saudemont, Medium Principal Software Engineer, recently walked through how the team identified the problem and what they built to fix it. If you’d rather watch than read, you have two options: Watch a short version from Monster Scale Summit or an extended follow-up webinar

The feature store and its role in Medium’s recommendation system

The feature store ties it all together, ingesting user activity and internal events and feeding them to the ML models that power recommendations. It’s what enables customization like the “For You” feed that greets logged-in users.

Each feature is a property of an entity, usually a user or a story. Some are simple and static, like whether a user holds a paid membership. Others capture interaction history: which stories a user has read, what content they’ve recently been shown, etc.

The following diagram shows a highly simplified view of the Medium feature store architecture:

The problem with a relational features data model

When they built their feature store years ago, Medium used relational features for cross-entity relationships. Unlike regular features, a relational feature can have multiple values for a given entity ID. Each value is defined by a relation ID (the ID of the related entity) and a timestamp recording when the event occurred.

For example, a “story users have read” feature is attached to the story entity type. It relates to the user entity type, and its values indicate whether/when a given user has read that story. 

Andréas shared the following schema diagram to explain the concept:

Features sit at the center, each attached to an entity type and defined by name, version, and data type. Non-relational features are simply a feature, an entity ID, and a value. Relational features add a relation ID mapping to another entity type, plus the value itself and a timestamp.

This approach proved suboptimal from a data modeling perspective. Since relational features link two entity types, the data ends up split between two tables: one for the entity IDs and one for the values. That means you can’t get both in a single query. The first query retrieves only entity IDs (not their associated values) and relies on ALLOW FILTERING. A second query then runs for each entity ID to fetch its value. “If we have 1000 entity IDs for which we want to fetch values, then we have to run 1000 queries to fetch these values,” Andréas said.

Overrelying on ALLOW FILTERING made things worse. “This is bad,” Andréas said, referring to monitoring data showing that 90% of rows read via these queries were simply discarded. “This is just data that we don’t need. ALLOW_FILTERING should be an escape hatch, not our design pattern.”

“ALLOW_FILTERING should be an escape hatch, not our design pattern.”

The list feature model

So they reinvented their data model and shifted to a list-based feature model. Instead of splitting data across two tables, everything for a given entity lives in one place and is retrieved in a single query.

Like other features, a list feature is defined by its entity type, name, and optional version. What’s different is the value. While a non-relational feature has a single value, such as true or false, a list feature’s value is a collection of items, each containing a value and a timestamp. Item values can be of any data type; the feature store doesn’t enforce consistency within a list.

For example, consider a user’s reading history. The entity is user, the feature name is reading history, the TTL is 6 months. After that TTL is reached, the data is automatically dropped by the database (since older history isn’t useful for recommendations). The list for a given user is a collection of story IDs and the timestamps at which they were read. The same story can appear multiple times, and multiple items can share the same timestamp.

A range of operations need to be supported. Create List and Delete List operations run at most a few times per day. Remove List Items with Value, which lets a reader scrub a specific story from their history so it stops influencing recommendations, runs at 1k-10k per second. Add List Items is higher still: every story read and every thumbnail shown to a user generates an event. Get List Items is the top, at 100k-1M operations per second.

“The Add List Items, and even more the Get List Items operations, are really the reasons why we need an efficient data store.”

“The Add List Items, and even more the Get List Items operations, are really the reasons why we need an efficient data store,” Andréas said.

Multiple items, one timestamp

Beyond raw efficiency, the new data model also had to support multiple items with the same timestamp. When Medium shows a user four story thumbnails simultaneously, all four presentation events share the same timestamp, but have distinct story IDs. If this isn’t handled correctly, primary key collisions occur.

The team’s solution was a single list_items table that stores everything.

The partition key combines feature_key and entity_id, keeping all items for a given list together. All of user 123’s reading history is stored in one partition, retrieved in one query. The clustering key concatenates each item’s timestamp with an MD5 hash of its value. The hash is what makes same-timestamp items with distinct values possible. 

Relying on MD5 hashes for uniqueness raises its own set of questions, but in practice, the team hasn’t seen collisions. “The values that we are storing are sufficiently distinct, especially when you add the timestamp into the equation,” Andréas said. The table’s clustering order is set to descending so ScyllaDB can optimize for the typical read pattern (most recent N items) rather than leaving the application to sort afterward.

TTL to control storage costs

Storage cost is controlled entirely through ScyllaDB’s native TTL, with no cleanup logic required. Every row expires automatically based on its own timestamp plus the feature’s TTL duration. “We don’t have anything to do regarding that,” Andréas said. “Any row for which the TTL is expired will be considered deleted by ScyllaDB.” 

Storage plateaus for a steady write rate. When a feature is retired, its data drains away on its own. “That’s super useful for controlling our storage and usage costs.”

Implementing the list operations

Add List Items is a logged batch of INSERTs with atomicity guaranteed: all items land or none do. Each row carries its own TTL calculated from its timestamp, so older items expire sooner. Since items almost always carry a current timestamp, new entries append to the top of the partition, which is exactly where reads will look first.

Get List Items runs as a single-partition SELECT with a minimum timestamp and a row limit. “We run the query on a single partition,” Andréas said. “That’s the maximum efficiency that we can have.” The clustering key handles filtering and ordering directly. Post-processing is not required.

Remove List Items with Value is the one operation that couldn’t be reduced to a single query. Because value isn’t part of the primary key, a direct filter isn’t feasible.

A local secondary index built specifically for this case first finds the matching item keys, then a batch DELETE removes them by primary keys.

“Using an index is really faster than a scan because the query is highly selective,” Andréas explained. “We have very few items in a given list that have the same values compared to the total number of items in a list. And thanks to the current structure, using a local secondary index is faster than a global index.”

Andréas shared another example. Starting with the original table partition, the goal is to delete all items with the value “storyC.” Using the local secondary index, the system first identifies the two rows containing that value. It then issues two DELETE statements using the item keys from those rows, which removes them from the list. The final operation, removing all list items, is even more straightforward.

“We can just drop the partition,” Andréas said, “and ScyllaDB does its magic. It just deletes all the rows for that partition, which means that it deletes all the items for the given list. And bonus point: it’s atomic. It’s either completing successfully or not changing anything at all.

ScyllaDB vs. DynamoDB performance

Medium implemented the list operations on top of both ScyllaDB and DynamoDB. The main goal was to benchmark how both databases compared on their actual production data. “Conceptually they are very close,” Andréas noted, “but they have significant differences in how they operate.”

For AddListItems, P50 latencies were low with both databases: ScyllaDB came in under 1.5ms, DynamoDB under 5ms. “DynamoDB is extremely fast, not as fast as ScyllaDB, but extremely fast at sub 5ms latency,” Andréas commented. Things got more interesting at the P95 and P99 latencies. ScyllaDB held steady at around 5-6 ms P95s, while DynamoDB ranged from 13-45 ms. ScyllaDB’s P99s were steady single-digit milliseconds, while DynamoDB’s ranged from 40- 120 ms.

It was a similar story for GetListItems. At P50, ScyllaDB clocked in at 1 ms, DynamoDB at around 3.5 ms. At P95, ScyllaDB held around 5-6 ms while DynamoDB spiked from 30 – 60ms. And at P99, ScyllaDB remained at ~30ms while DynamoDB ranged from 70 ms all the way up to 220 ms.

“ScyllaDB is very fast, with very predictable performance, and that’s super important for us.”

One caveat: DynamoDB was running without an extra caching layer. “We expect that could have a significant impact for DynamoDB because of the high cache hit rate that we are seeing on the list,” Andréas said. “But we don’t have the data yet, so we cannot compare them.” His verdict for now: “ScyllaDB is very fast, with very predictable performance, and that’s super important for us.”

Key takeaways

One pleasant side effect of getting the data model right: Medium is now eager to use ScyllaDB for additional feature store workloads. Before, they were holding back because they didn’t want to build on the shaky relational feature foundation.

Reflecting on the path to this point, Andréas left the audience with this parting advice:

“If you have a suboptimal data model, you will have queries that are slow, that will scale badly. And most likely, you won’t be able to optimize that data model. You will have to define a new data model that will be better. So take time to think about your data model before you start the implementation, because once you have production data using your suboptimal data model, it’s too late.”

The post When your data model is the bottleneck: lessons from Medium’s feature store appeared first on The New Stack.