A Monday with three topics that deserve the attention of anyone building products with AI: pricing, tool security and the effect of AI on software maintenance.

Every AI product has a decision hidden behind the price: how much usage fits inside the plan.

Today I read an account from a developer who used ChatGPT’s unlimited chats as a small programming team. They opened several conversations at once, each handling one task, and barely touched the Codex limits. According to them, OpenAI announced that unlimited chats will end and that this usage will be unified with Codex.

I don’t know the details of the change, and the text doesn’t give dates either. What caught my attention is the pattern. When a plan promises unlimited usage, the people who benefit most are precisely the ones who consume the most, and those users find ways of using it that nobody anticipated in the original design.

On the product side, this raises three questions I like to ask early. Who is the user who consumes my product the most, and how much do they cost to serve? Is the limit clear to the customer, or do they only find out when they hit it? And if I need to change the plan, how do I tell people without breaking the trust of those who built a routine on top of it? The guide on AI agent cost management helps you start from cost per task.

I think it’s healthy that the market is maturing on this. AI pricing is product, and it deserves the same discovery care we give to any feature.

For those who are curious and want to read the full account, here is the link.

The rest of the radar

  • Backdoor in the Claude Code extension for VS Code on the Cursor Marketplace — an agent tool installed through a marketplace is a supply chain; versions 2.1.312 and 2.1.313 were removed, and uninstalling alone is not enough. Have an incident response plan for cases like this. Phorion.
  • Postgres 19 is late because of LLMs — AI finding bugs at scale makes maintenance more expensive; backported commits rose more than 60%, according to the author. This belongs in the AI governance of anyone who depends on open source software. Robert Haas.

That’s it for now. More tomorrow.