The need
A user reports Codex has been acting 'spaced-out': looping, saying it will do something then not doing it — and is falling back to Gemini/Claude case by case with no signal of when each agent is reliable.
Original post — verbatim Thai
ทำงานแบบเอ๋อๆ พาวน บางทียอกจะทำแล้วก็ไม่ทำ กลับมาซบอกพี่geminiกับclaude
Same pain, other posts
อยากเห็นเคสที่พังจริง ๆ ว่าแต่ละคนเจออะไรกันมาบ้างครับ
Three business opportunities
OPPORTUNITY 01
Agent health monitor
- What
- A CLI wrapper that logs runs across coding agents, scores completion rate and detects loops/stall patterns, publishing 'which agent is solid today' guidance.
- Who pays
- Professional devs and small studios pay a subscription for routing confidence.
- Price
- ฿149–399/mo — hypothesis
- Why now
- Multi-agent fatigue is visible across this group's posts; nobody aggregates reliability telemetry.
- Effort
- M
- Moat
- A longitudinal cross-agent eval dataset that compounds.
- Risk
- Users may tolerate free heuristics; per-agent telemetry is fragile across releases.
OPPORTUNITY 02
Auto-routing agent gateway
- What
- Proxy that dispatches coding tasks to Codex/Claude/Gemini by task type with retry, loop detection and guardrails.
- Who pays
- Teams pay per seat for higher throughput and fewer dead loops.
- Price
- ฿99–299/seat/mo — hypothesis
- Why now
- The poster's manual model-hopping is exactly what routing automates.
- Effort
- M
- Moat
- Routing telemetry that continuously improves per-week winner picks.
- Risk
- Vendor terms and API churn; provider bundles ship the routing.
OPPORTUNITY 03
Agent reliability runbook workshop
- What
- Training for Thai teams on eval gates, checkpoints and agent QA rituals; includes installing the monitor.
- Who pays
- Companies pay per seat for risk-managed agent adoption.
- Price
- ฿2,000–6,000/seat — hypothesis
- Why now
- Teams adopting agents hit the reliability wall the post describes.
- Effort
- S
- Moat
- Case library from real Thai deployments.
- Risk
- Niche willingness to pay for process training.