Trying to buy a book on Amazon. Last one in stock. Checkout. Payment denied. Trying to buy again with another card: book not available.
Software Engineering is not solved.
“This affects almost no users. Only millions.” is something you will hear a lot in companies like Google, but it will usually be in % terms so “only millions” doesn’t echo on the walls of the meeting rooms.
Widespread voter suppression is reportedly being carried out by local government officials aligned with Flávio Bolsonaro. They're breaking the law by keeping public buses in the garages to prevent poor and working classes Brazilians from voting. This is very serious.
Major cities suppressed voting by keeping public transport buses out of circulation. Salvador-BA as a whole and in São Paulo (city), selected voting areas where Lula did well in the first round. Heavy rain in the South, where Bolsonar0 does well, has also reduced access.
Reports on Brazilian TV of terrible chaos. Polling stations that had already closed their doors, reopening. Poll clerks completely confused. What a nightmare. Hard to say what the consequences might be, tonight and in the weeks and months ahead.
How long do we wait before we declare bugpocalypse over/not the end of the world? Lights are still on. Banks still working. Water, gas, food, hospitals seem functional so far. When do we decide we’re out of the woods? Weeks? Months?
Pessimistic concurrency control (e.g. a merge queue) is required when workers in a concurrent system start doing more work. Optimistic CC assumes (optimistically) that conflicts will be rare.
Has anyone else noticed that a merge queue becomes effectively required with agents. I'm not totally sure how to address this yet. But you can't have a stampeding herd of agents rebasing each time something is merged. I'm getting really good at wasting money on crap like this.
@dcolascione They probably have used it, but fuzzers are not exhaustive. People that say C/C++ safety is solved because of ASan and TSan underestimate the size of the search space.
> Their advice is to fold existing subagents into the main loop, turn their instructions into skills, expose their actions as tools, and run evals to see if performance holds up.
This is what anyone with any taste in task runner design knew all along.
Anthropic is now recommending that companies REMOVE unnecessary subagents from their AI systems.
This is a pretty significant shift in agent architecture, with their team finding that a single agent equipped with skills can frequently outperform complex multi-agent systems, and one customer seeing roughly 2x faster execution.
The emerging rules of thumb:
→ Skills: Teach the agent HOW to do something, including procedures, expertise, and instructions.
→ Tools / MCP: Give the agent the ability to DO things through APIs, actions, and live data.
→ Subagents: Use for parallel work, isolated context, or tasks a single agent struggles with.
The interesting part is progressive disclosure. Instead of spawning specialized agents or stuffing everything into a giant prompt, one agent discovers and loads skills and tools as needed.
Their advice is to fold existing subagents into the main loop, turn their instructions into skills, expose their actions as tools, and run evals to see if performance holds up.
As models have gotten smarter, improving agent architecture is becoming less about what you can add and more about what you can delete.
https://t.co/DNq60Skkmc
You have skills (extra context) and tool calls as a black-box task dispatching mechanism, why not fold the spawning of agents into the same uniform concept?