A Saturday with a team bet at Cisco, a backdoor in an AI extension for developers and two pieces on what automation does to the skills of the people who use it.

Some product decisions don’t look like product decisions. Giving unlimited access to a tool, for example.

Cisco did exactly that with a 27-year-old engineer. According to Fortune, Jeetu Patel, the company’s president and chief product officer, gave unlimited AI tokens to Paddy Krishnamurthy, who joined Cisco through the acquisition of a startup. The task is to rewrite legacy code, with a goal of refactoring 98% of it in 24 months. He has four weeks to prove it works.

From a product point of view, what stands out is not the “unlimited”. It is that it only exists because there is a clear hypothesis and a short deadline to validate it. A budget with no ceiling for someone who knows what they want to test is an experiment. For everyone else, it is just a cost. The guide on AI agent cost management helps put this kind of bet into the math.

Patel also talked about people. He says he found this engineer five levels below him, in a job that was not very strategic, and that he wants young people teaching the more experienced ones, so-called reverse mentors. The report cites a survey saying that about 40% of employers plan to hire fewer recent graduates because AI does the work more cheaply.

I think this conversation matters. A good AI team is rarely made only of specialists or only of digital natives. Patel describes a third of experienced operators, a third of systems thinkers and a third of AI-native people. It is not a recipe, but it is a team design worth discussing. This is part of AI product management.

Since a good part of this comes from an interview, the result is yet to be proven. Four weeks is not much time, and it will be interesting to see what comes out.

If you are curious and want to read the full report, here is the link.

The rest of the radar

  • Backdoor in the Claude Code extension on the Cursor Marketplace — AI tools for developers have become an attack vector; it is worth having an extension approval policy and an incident response plan (versions 2.1.312 and 2.1.313 affected). Phorion.
  • AI and the loss of SRE practice — if AI handles the routine, humans lose the skill for the rare and serious; design the product to keep people engaged, not just supervising. SigNoz.
  • Sleep framework (Google Research) — a method for LLMs to consolidate what they learn without forgetting what they already knew; it is still research, with no product or timeline, but it points to models that learn after launch. Google Research.

Next week there is more, with the same idea: separating what changes the product decision from what is just noise.