Welcome to Runtime! Today on Product Saturday: Honeycomb rolls out new AI agents for individuals and teams trying to keep sites online, IBM and AMD's latest attempt to keep quantum computers honest, and more.
Please forward this email to a friend or colleague! If it was forwarded to you, sign up here to get Runtime for free every week, or level up here.
Ship it
Worker bees: Nearly a year into the AI coding agent explosion, enterprise tech has still not quite figured out how to deal with the massive amount of new code flooding into their repositories. Some like Microsoft's Brendan Burns think that a lot of code will eventually become "transient," or invisible to developers the same way nobody really understands what assembly code is doing, but we're not quite there yet.
Honeycomb basically defined the modern "observability" movement back when it launched a decade ago, and the concept has never been more important as companies struggle to understand what their agents are doing and manage that activity. This week it introduced a new set of AI agents for site-reliability engineers tasked with making sure all that new code doesn't take down production.
The new agents fall into two different categories: Personal Agents, which were designed to be used individually to help SREs diagnose problems and find solutions, and Team Agents, which can take actions as defined and managed by a group. "Give the agent a goal, such as monitoring deploys, and it works in the background on a schedule, trigger, or service level objective (SLO)," Honeycomb said in a blog post.
No two production environments are exactly alike, which means agents trained on the data that emerges from that operating environment could start to quickly understand what went wrong, why it went wrong, and take informed actions to get back up and running. Ideally that information could filter back into the coding agents themselves to prevent future incidents, and that could make admins feel a little better about this next chapter of the automated enterprise, which is coming one way or another.
Delivery, continued
The error era: One of the many reasons why quantum computing has been five years away from making an impact on the enterprise for the last 20-odd years is that researchers have yet to figure out the best way to prevent quantum computers from making errors, which is more or less a solved problem in classical computing. This week IBM, AMD, and Oak Ridge Laboratory introduced "a new co-design platform for quantum error correction" that relies on modern high-performance computing clusters for an assist.
"This co-design platform will allow researchers to evaluate the tight integration of classical and quantum computing to fuel momentum in quantum error correction research," the companies said in a joint press release. The platform combines IBM's Quantum Nighthawk processor with a supercomputer cluster running on AMD's CPUs and GPUs to "accelerate progress toward a scientifically relevant quantum computer by 2028," although the timeline for a commercially relevant quantum computing remains a moving target, probably somewhere around five years away.
Context clues: Snowflake spent a lot of time at its annual customer conference last June talking about "context," the notion that AI agents need to understand relevant business information — not just hard data — if enterprises are going to see results. This week it added new features to Horizon Context, the service it introduced in June to connect agents and business intelligence data.
"Snowflake’s new Business Ontology allows these agents to operate with a shared, pre-approved understanding of core business definitions to enable reliable answers," the company said in a press release. The idea is that "every business definition needs an official version and an accountable owner, whether the number is going to a team meeting or to a regulator," and Business Ontology will allow companies to enforce those policies for both humans and agents.
Build the foundation: While context is indeed important, AI agents still need to understand where and how to find the raw data they need to accomplish their tasks, and just as SaaS applications are being redesigned with agents in mind, so are databases. This week MariaDB launched the latest version of its flagship database that promises improved performance and reliability as well as new and better ways to work with agents.
The MariaDB Enterprise Platform 2027 release adds features such as AI Retrieval, which translates natural-language commands into SQL, and Agent Memory, which improves the reliability of long-running agents. "Most enterprises ... will add agents to the systems that already run the business, which makes the transactional core and its ability to serve analytics, retrieval, and memory from the same live data the practical starting point for agentic workloads," IDC's Devin Pratt said in MariaDB's press release.
The rest of The Stack
Find your way: Model routing looks poised to become an important part of the enterprise AI stack, and Stripe's acquisition of OpenRouter for $7 billion underscores the strategic importance of matching the right application with the right model. Companies that prefer an open-source approach might want to take a look at Agent Router, which was based on the popular Envoy gateway project.
Mary Branscombe
Buildings, blocked: The seismic results of Tuesday's elections in the US will probably change the way several states approve new data-center projects thanks to skeptical governors and voter-approved regulations. The AI industry, and its investors, are worried.
Yaël Bizouati-Kennedy
Thanks to everyone who attended the Stack Summit in London this week! Stay tuned for more details about gatherings planned for 2027!
Quote of the week
"When you're with the hyperscalers, you [can] have a completely different level of resilience. You often have six networks between different data centers, and someone else is managing it for you. If we need to ramp something up and ramp it back down again, we can. You need to take the effort to actually do that, or the costs can explode. But from an operational point of view? No regret on that move." — Nestle's Group Chief Data and Information Officer Chris Wright, explaining to The Stack how a move to the cloud several years ago set the company up for the AI era.
We're also reading:
It's time for frontier LLMs to take a responsible approach to Cyber: Palo Alto Networks CEO Nikesh Arora put the model companies on blast in an essay posted to X arguing they need to take a more "collaborative" approach to vulnerability management.
Deprecation Notice: Redis engine for Elasticache: The federal government is moving off Redis and on to Valkey, which is interesting.
Compass Point: Azure Capacity Checklist, Agent Teammates, Sovereign Cloud: Directions on Microsoft discusses several issues that Microsoft Azure customers will want to read, including the return of Azure capacity issues.
Thanks for reading — see you Tuesday!
