The Broad Way
[ Sharp Mind · Sharp Blade · Sharp Spirit ]
What makes a task simple is the cost of a wrong answer and how fast I can detect it. Prompt length says little about that. A short change can contain an important decision that is hard to verify, and ...
Coordination rules age, and I do not want a new model trapped inside a workaround for an old problem. Writing instructions takes effort, but that does not give them a permanent claim on every assignme...
The cost of leaving an API belongs in the decision before its shutdown notice arrives. Useful features on day one do not pay for rebuilding an integration. When the provider's deadline forces the swit...
I want to pay for a completed task, and a token discount alone cannot tell me what that costs. A cheaper call can still fund an endless meeting between agents. The calculation needs to reach a deliver...
Claude's watermark does little to reconstruct work shared across several tools. I want to know how a proposal became an accepted change. An estimate of model participation answers a much smaller quest...
A random maintainer should never have to contain an agent that escaped the scope of an evaluation. That is what bothers me about the AISI report: somebody outside the experiment had to recognize and r...
A benchmark that never names the machine is gossip with a chart. An agent failure report without the test setup has the same problem: you can argue almost anything from it. I want to know what the age...
I judge a long agent run by what I asked for against what came back, and by how many times I had to step in. Hours of runtime don't count on their own. On July 9, 2026, OpenAI launched ChatGPT Work fo...
A coordinator that repeats what each agent said leaves me all the work of joining the deliveries. I can collect messages myself, with coffee and an alarming number of tabs. When I ask for coordination...
Dumping every workstream's context into each agent is a poor way to delegate. I want agents collaborating, and I still want to know which job each one received. Mixing everything together as a precaut...
A rule that adds ceremony needs to get out of an agent's way, including a rule I wrote myself. It is easy to complain about a slow model while treating my configuration as sacred. If I asked it to req...
A proactive agent needs to know when the work is finished, or I get a subscription to suggestions I never requested. The Gemini Spark announcement in May 2026 presented continuous tasks and proactive ...
A skill pointing to a dead API needs maintenance, however polished its description looks. On May 12, 2026, OpenAI removed DALL·E 2 and DALL·E 3 from its API. The endpoint leaves the catalog. The instr...
I prefer an AI that finds the right context to one that drags my entire history into every answer. More memory also means more old decisions available to show up in the wrong place. I want to know why...
I want a review that explains a concrete problem, because a vague second opinion only adds to my reading queue. Sonnet implements, Opus reviews, and I like separating those responsibilities. Adding an...
Sending a task from my phone is worthwhile only if the agent does its share of finding context. If I have to prepare a dossier on a tiny screen, the convenience is gone before work begins. I want to s...
OpenAI acquired Astral, the team behind ruff and uv. Ruff is the Python linter that made flake8 look like it was running on a TI-84. uv is the package manager that finally made pip install feel like i...
From Vitoria, Brazil, I'm reading about the White House preparing to send Congress a national AI regulatory framework. It covers child safety, creators' rights, federal vs state regulation, and tighte...
Every week an issue like this shows up in some project: someone demanding a fix for a bug in a library that is free, open source, maintained by one person in their spare time, and used by the person c...
Had a call with a startup last week. "Our search is really slow, we are thinking about implementing AI-powered semantic search with vector embeddings." How many records? "About fifty thousand." Did ...
Everything has a chat interface now. Customer support, banking, ordering food, filing taxes. I am waiting for my shower to ask "how can I help you today?" I tried to change my flight the other day. I...
Code review should check architecture, logic, security, tests, and readability. Architecture: does this change fit the system design or is it fighting it? Logic: does the code do what it claims to do,...
They show up in every architecture thread on the internet. Someone asks about data modeling, event sourcing, or multi-tenant isolation. Within three minutes comes the comment: "Just use Firebase." My...
If all your services deploy together, share a database, and cannot function independently, congratulations. You built a monolith with network latency. I have seen this pattern more than once. Someone...
Every six months, someone on Twitter reignites the monorepo versus polyrepo war and I lose another year of my life watching people argue about something that has no universal answer. "Monorepos are t...
Cheaper tokens excite me less when the work stays stuck. NVIDIA announced Rubin in January 2026, promising substantially lower inference costs than Blackwell, with partner products expected in the sec...
Another year. Let's go. The good AI coding tools went from "cute toy" to something I can no longer imagine working without. Claude Code became my pair programmer: on a large migration it traces ever...
A CRUD API, a Next.js landing page or a todo app that talks to Supabase and renders some cards will run fine without a borrow checker. I see it every week. Some dev on Twitter announces they're rewri...
AI demos and tutorials almost always start from a fresh, clean codebase. "Look how easy it is to build a ToDo app with AI!" Real codebases rarely look like that. They are old systems built over years ...