Eugene Oleinik

Founder in Residence @AntlerGlobal. Ex-CTO 12Go (10yr). Building tools for developers. Singapore.

[Eugene Oleinik @EugeneOleinik]

* don't forget to stay focused on the most important tasks, as always

[Eugene Oleinik @EugeneOleinik]

* everybody is using coding agents
* how to orchestrate team development around this?
* keep updating the knowledge base and guidelines (it's easy to automate) (we use /curate-all skill after each feature that updates docs, tests, dashboards and skills)

[Eugene Oleinik @EugeneOleinik]

* love mechanical verifiers - AI has amnesia, so put as many gates as you can, it's cheap to write while the context is fresh. they can be in GitHub CI
* code review is Developer A's AI talking to Developer B's AI

[Eugene Oleinik @EugeneOleinik]

Published our Agent Experience principles as an installable skill. The installer violates 3 of the 11 principles it distributes.

Agent-first is mobile-first for developer tools.

[Eugene Oleinik @EugeneOleinik]

Chak-chak (Tatar): "looks like deep-fried worms"

Chakku-chakku (Japanese rebrand): "artisanal tempura crisps, added to my bucket list!"

Same dessert. Different story.

Your product doesn't need to change. Your positioning does.

[Eugene Oleinik @EugeneOleinik]

Vercel now serves markdown to agents via Accept header. 20kb → 2kb.

CSS, JS, frameworks becoming optional for AI consumers.

Try it:
curl -H 'Accept: text/markdown' https://t.co/YV1gQiYcUX

https://t.co/Igu4X6qI19

[Eugene Oleinik @EugeneOleinik]

I taught Claude 10+ skills. It ignored them all.

The fix: learnings/ folder + post-commit hook.

Skills = workflows you invoke
Learnings = facts Claude searches

Works 4/5 times vs ~1/10 with skills.

https://t.co/IzhwfgliWO

[Eugene Oleinik @EugeneOleinik]

I taught Claude 10+ skills. It ignored them all.

The fix: learnings/ folder + post-commit hook.

Skills = workflows you invoke
Learnings = facts Claude searches

Works 4/5 times vs ~1/10 with skills.

https://t.co/IzhwfgliWO

[Eugene Oleinik @EugeneOleinik]

Two years ago I asked my CEO if we should use ChatGPT. "It leaks everything," I said. His answer: "Everyone's using it. If we don't, we're behind."

He was right. You'll keep using these tools. So will everyone else. That's the trap.

[Eugene Oleinik @EugeneOleinik]

My https://t.co/sHVJPM2BrS hit 55KB. Sessions got short, compaction constant.

The fix: split into learnings/ folder. Let the agent grep and cat like it navigates code.

55KB → 24KB. All learnings preserved. Sessions run longer.

Filesystem + bash > custom tooling.

[Eugene Oleinik @EugeneOleinik]

Friends debating "is AI a bubble?" Zero data, just hot takes.

Polymarket has $305k in bets:
→ Burst by 2025: <1%
→ By 2026: 30%

Skip the debate. Check what people with money on the line actually think.

https://t.co/cMRPmyPZqP

[Eugene Oleinik @EugeneOleinik]

git checkout -b feat/x → Neon DB branch created
git checkout main → orphan branches auto-deleted

Add "never work on main" to https://t.co/sHVJPM2BrS and your AI assistant can safely run migrations without touching production.

Zero-friction database branching ftw

[Eugene Oleinik @EugeneOleinik]

ssh server -t tmux -CC — one flag I missed for 10 years. Native Mac tabs for remote tmux.

https://t.co/mWkseDoYzI

#tmux #macOS

[Eugene Oleinik @EugeneOleinik]

Chat AI has no survival instinct. Robots will. That's not alarmism—it's an engineering requirement.

https://t.co/T1IHcRpvS5

#AI #Safety

[Eugene Oleinik @EugeneOleinik]

Server kept rebooting. No logs. Claude Code found it: PSU hitting 750W limit.

https://t.co/GQ1JynhQbs

#ClaudeCode #Debugging

[Eugene Oleinik @EugeneOleinik]

87 lines of bash → reusable GitHub Action. AI-powered changelog summaries to Slack.

https://t.co/rzf1mZ060P

#GitHub #Automation

[Eugene Oleinik @EugeneOleinik]

Your AI coding agent has amnesia. https://t.co/sHVJPM2BrS is the fix—persistent context across sessions.

https://t.co/kL84PHnBto

#ClaudeCode #AI

[Eugene Oleinik @EugeneOleinik]

Voice-to-text transcribes exactly what you say. That's the problem. LLM post-processing cleans it up.

https://t.co/MqhxwfFjpK

#VoiceToText #LLM

[Eugene Oleinik @EugeneOleinik]

The AI iteration trap is really a clarity trap.

You iterate because you don't know what you want.

The fix isn't "read more" - it's defining acceptance criteria before you start.

If you can't answer "what would make this obviously acceptable?" - you're not ready to iterate. ...