Pinned Tweet
Most AI agents work in the demo and die in production.
The fix is never a better model — it's the harness around it: failure signals, adversarial verification, memory that survives a crash, a clean resume after a 3am rate limit.
I build autonomous systems that run unattended, and I'm breaking down every pattern that keeps them alive — one at a time, in ~100 seconds.
Faceless. Build-in-public. Open code → github.com/alkas79
Follow if you ship agents.
Hitting a rate limit mid-task isn't an if. It's a calendar entry. Checkpoint after every step that costs something, and make 'resume' the same code path as 'start.' If resume is a special case, it's the one that breaks.
3
1
134
Your request timed out. The server didn't. Retry that blind and your agent posts it twice, charges the card twice, emails the customer twice.
62
'Tests pass' from the same agent that wrote the tests is a closed loop with a press release. Hand the claim to a fresh agent with no shared context and one job: prove it false. A surprising amount of 'done' dies right there.
1
1
57
Nine days of green checkmarks. The login behind them expired on day one. Nothing crashed, so nothing alarmed. Now every job has to prove it produced something, or it fails loud.
2
1
109
'Done, all tests pass' is the agent's press release. The log tail is the audit. Read the last 50 lines before you believe a word of the summary.
1
1
89
Sort every agent action into two piles. If you can undo it in a minute, automate it. If it sends, deletes, deploys or pays, a human clicks confirm. Most of the autonomy debate goes away once you draw that line.
3
3
101
Zero rows from a step that usually returns forty isn't a quiet day. It's dropped work, and it passes every check you have. Give 'nothing to do' and 'something broke' different exit codes, or a customer will be the one who tells you which it was.
39
A verification step costs a few seconds and a few cents per run. The bug it would have caught costs a customer email, a backfill script and a week of someone's attention. Pay at the gate or pay at the incident.
1
35
Least privilege for agents isn't paranoia. It's the difference between 'it mangled one draft' and 'it mangled the CRM.' Same bug. Very different Monday.
1
47
Can you rerun last Tuesday's failure from its logs alone? Same inputs, same tool responses, same model output? If not, you're not debugging, you're writing fan fiction about what probably happened. Log every model call and tool result.
1
49
95% accurate sounds great until you ask what the other 5% does. If a wrong call means a reply someone ignores, ship it. If it means a wire transfer, hold it for a human. The price of being wrong sets how much you automate, not the accuracy.
53
Count how many times your agent said 'I'm not sure' last month. If the answer is zero, you built a bluffer. Make abstaining a valid output and score it above a wrong answer.
2
2
74
Alexandros Karras retweeted
Today, we’re bringing GPT-Image-2.6 to Codex and the API.
Image-2.6 Flare and Sunburst, our two new API image models, generate images 50% faster than before.
At medium or high quality, Sunburst keeps details more consistent from one edit to the next.
Cron doesn't leave a goodbye note. The job just stops showing up, and every dashboard stays green, because green is what nothing looks like. Alert on the missing run, not just the failed one.
8
4
128
Git had my agent's work. Three staged files, never committed. A cleanup command removed the only thing that pointed at them. We got the contents back. Their original file paths were not recovered.
1
55
Staged files in a worktree stay alive through its index. That folder had gone missing. My agent removed 12 finished worktrees, then ran git worktree prune, which also deletes a missing worktree's index. The bytes were still in git, so we pinned them to a rescue ref.
1
20
That ref lived only on this machine. My backup assumed git restores from GitHub. So it went to a second drive as a verified git bundle. Prove you can restore what your agents create. Free: The Reliable Agent Handbook — 10 patterns, 20 min → buttondown.com/alkasbuilds?u…
22
Someone tweaked one line of the system prompt to fix the tone. Three unrelated flows quietly started skipping a validation step, and nobody noticed for a week. Fifty real cases run on every prompt change would have caught it before lunch.
1
1
49
My repair reached round 17 and was still going. It began as display fixes, so my status page would never show all clear over data it hadn't read.
2
4
78
After round 3, full proof was the bar instead of a narrower fix. Rounds 4 to 17 took under a day. The change grew from 48 files to 72. Several reviews found the new hole in code the last repair had added. Round 8 added auto-recovery that could itself print "All lanes fresh".
1
17
A stop rule set during round 17 ended it. One more check, then ship. The rest became 14 named follow-ups, and the follow-up got its own stop rule on day one. Set yours before round one. Free: The Reliable Agent Handbook — 10 patterns, 20 min → buttondown.com/alkasbuilds?u…
11

