Agents fail quietly without context, harness, and human gates
Pavan_Belagatti · x · 2026-11-15
The author argues that creating an agent is the easy part — what makes agents useful in engineering is everything around them:
- Context: agents need knowledge of your services, teams, deployments, and incidents, or letting them deploy code ends in failure
- Harness: tools (MCP/actions/APIs), memory (state across runs), skills (prompts, golden paths), plus a continuous plan→act→observe→evaluate loop
- Human gates: approval checkpoints for plan sign-off, PR review, and production promotion
- SDLC integrations: wiring into existing Plan/Code/Build/Secure/Ship/Operate toolchains
The attached diagram lays out a full Agentic Engineering System architecture combining these layers.
More from coding & agent
- Cursor runs unattended for 1+ hours: watching processes and auto-fixing failures — NickPassig · 2026-11-15
- Stanford's CS329A Self-Improving AI Agents course: 9 free lectures now online — omarsar0 · 2026-11-15
- Flex-RSI benchmark tests whether agents can self-improve across 18 frontier models — _vztu · 2026-11-15
- Google open-sources AX, a Kubernetes-style orchestrator for running fleets of autonomous agents — rakyll · 2026-11-15
- Reddit: swapping prompt tweaking for four context files made Claude output dramatically better — Exact_Proposal8849 · 2026-11-15
- Harrison Chase on goal design: as agents work on goals, designing them well matters — hwchase17 · 2026-11-15