Heya and welcome back to Five Things AI!
We spent two years being told AI would change everything. This week we get to read the receipts. Enterprises are throttling usage because the tokens are too pricey. Architectural bloat does not beat one strong agent once cost enters the math. Google quietly drops a boring file format that might matter more than any of it. And the economists measuring the whole shift cannot agree on the basics. The technology is maturing. So are the questions about what it is worth.
Read on, my dear! Or have your agents tell you what to read first!
Companies Are Throttling Employees’ AI Use Because It’s Too Expensive
The news shows the looming fallout from companies adopting AI as quickly as possible, and AI providers’ moves to charge enterprises based on how much they use AI rather than a flat fee. Emails obtained by 404 Media even show some companies cutting off access to some AI models altogether in an attempt to stop burning through their AI tokens, and big tech companies like Adobe are ending unlimited access to Claude.
We are currently in a stage where we have to figure out what works and what doesn’t. As always, some companies moved faster than others and then had to recalibrate their efforts. It is a sign of the maturing of the industry, not a sign for concern.
Open Knowledge Format (OKF): The Markdown Standard Built for AI Agents
The agent boom has created a strange bottleneck. Models are getting better. Tool-calling is getting more reliable. Protocols like MCP are making it easier for agents to connect to external systems. But the knowledge those agents need — what a metric actually means, which table joins to which, how an internal process works, why a dashboard number changed last quarter — is still scattered across wikis, data catalogs, shared drives, code comments, Slack archaeology, and the heads of senior employees.
Google’s bet with OKF is simple: before every company builds yet another proprietary knowledge layer for agents, maybe we should agree on a boring file format first.
Granted, this is still early with OKF 0.1 just being released, but when they did, I was kind of suprised that not too many people took note. Google has developed and released quite a few standards around interoperability in the Agentic AI space and hopefully they will hand this over to the Agentic AI Foundation as they did before with the A2A Protocol stack.
(…continue reading.)
Smooth AI criminal drives ‘first’ end-to-end agentic ransomware attack
JadePuffer’s “self-narrating” payloads “contained natural language reasoning, target prioritization, and the kind of detailed annotations that human operators don’t often write but LLM-generated code produces reflexively,” Clark added. “The operation also adapted in real time, retrying failed steps within refined parameters. In one sequence, it went from a failed login to a working fix in 31 seconds.”
After exploiting CVE-2025-3248, a missing authentication vulnerability in Langflow that allows remote, unauthenticated attackers to execute arbitrary Python on the host, the AI agent began scanning for and collecting secrets, including LLM provider API keys, cloud credentials “with explicit coverage of Chinese providers” including Alibaba, Aliyun, Tencent, and Huawei, while also scanning for AWS, Azure and Google Cloud Platform, cryptocurrency wallets, and database credentials.
What a time to be alive, we now have agentic criminals.
Multi-agent AI keeps collapsing back into one agent. A fair test just proved it.
They ran the audit across a stratified set of backbone models (GPT-4o, GPT-5, GPT-OSS 120B, and Gemini-2.5-Pro) and a spread of hard benchmarks, including GPQA-Diamond, HLE-Math, SWE-Bench Lite, and a deep-research task called BrowseComp-Plus. The guiding question was simple: do these automatic systems deliver consistent, cost-effective gains over a strong single-agent baseline?
The answer was mostly no. Across the ecosystem, the automatically designed systems rarely outperformed a strong single agent once cost and baseline strength were accounted for. The authors describe this as evidence of “architectural bloat,” sophisticated agent graphs that do not translate into functional reasoning gains.
These findings are really interesting, yet counter-intuitive. I’d love to know how this would now look like with a bunch of agents running on cheap LLMs compare to Fable 5.
A.I. Is Reshaping the Economy. Good Luck Measuring How.
Researchers can’t even agree on basic questions like how many companies are using A.I. or which workers are most vulnerable to the disruptions it could cause.
The conflicting signals partly reflect the challenge of detecting economic shifts in real time. Government statistics are inherently backward looking, and they are better at measuring broad trends than developments in specific sectors or regions. New technologies that might lead to the emergence of new products, jobs or entire industries can be particularly difficult to measure.
Plus: AI comes in so many different flavors and usecases, the impact on each and every company when “using AI” can be totally different as well.
If you missed last week’s edition of Five Things AI, you can read it here:
That’s it for Five Things AI this week! 🤖
— Nico