Ai on jpcercal.com

Changing the Decision Model Changes the Policy

Sun, 11 Oct 2026 00:00:00 +0000

I’ve been reading Vercel’s comparison of Jev, Laya, and Liquid d1, and the useful part for me is not which model wins.

The article does not establish a winner.

It shows that all three can answer the same typed questions through AI Gateway. That makes comparing them easier. It does not tell us whether the application should act on their answers in the same way.

That is the distinction I would keep before changing a model in an existing system.

GPT-6 Astra Is an Operator, Not a Chatbot

Thu, 24 Sep 2026 00:00:00 +0000

I’ve been thinking about GPT-6 Astra less as a new chatbot and more as a standing offer to operate the computer.

OpenAI calls it “the world’s most intelligent and aligned model.” Perhaps. The claim I can check is narrower. On the launch note and the later work note, Astra is described as something that uses the applications people already have open, including applications with no API.

That is the distinction I would use.

The Cost of the Loop Matters More Than the Score

Wed, 23 Sep 2026 18:00:00 +0000

I’ve been reading Anthropic’s notes on Claude Opus 5.5, and the part I keep coming back to is not the leaderboard.

They say it performs at the level of Claude Fable 5.1 on most work, and that in their own use the gap is narrower than the benchmark scores suggest. They also say that at this level of capability, benchmark margins have become a less reliable guide to real-world differences.

When AI Decisions Become Software Inputs

Wed, 23 Sep 2026 00:00:00 +0000

I’ve been thinking about what software actually needs when it asks an AI system where to send a request.

Usually it is not another paragraph. It is an answer it can inspect: a known category, an indication of uncertainty, and a rule for what happens next.

Consider a support message: “I was charged twice, and the receipt page will not load.” A free-form explanation might help a person, but the application still has to decide which queue receives the message. Parsing prose into a queue name adds another place for the workflow to fail.

What an AI Session Can Teach the Next One

Mon, 14 Sep 2026 00:00:00 +0000

An AI assistant can help write code, explain an unfamiliar system, investigate a failure, or draft documentation. An AI agent can go a step further and take several actions on our behalf.

In both cases, it is easy to focus only on the immediate result. Did the code compile? Did the document look reasonable? Did the task move forward?

There is another result that is easy to lose: what the session taught us about the way our team works.

Working With AI Agents: Stay on the Loop

Sun, 13 Sep 2026 00:00:00 +0000

I’ve been thinking recently about how our role as software engineers changes when AI agents become part of the development process.

There seems to be a tempting choice between two extremes.

Either we let the agent handle everything and trust the result, or we stay involved in every step, reviewing every piece of code it produces.

I don’t think either approach scales particularly well.

The output isn’t the whole process

When building software, we often focus heavily on the artifacts we produce: code, tests, technical designs, configuration, and documentation.