# Coding Agents Don't Scale Themselves. Neither Do Your Teams. — Patrick Debois, Tessl

Source: [AI Engineer](https://www.youtube.com/watch?v=zCJtYuqwm7E)  
Feed7 permalink: https://feed7.dev/p/coding-agents-don-t-scale-themselves-neither-do-your-teams-patrick-deboi-1tkeyjb  
Published: 2026-08-22T16:00:06.000Z  
Trust: Source Linked (source_linked)

## Why Included

Agent adoption becomes a team-systems problem: improve shared context and harnesses, assign platform ownership, and measure fewer human interventions instead of individual prompt speed.

## Source Summary

The talk frames adoption across team, platform, and organization. Teams improve shared context and harnesses; platform owners curate reusable skills, evals, guardrails, and identities; leadership gives those owners a mandate rather than relying on scattered experiments.

## Practical Implication

Move retrospectives from patching agent-written code toward fixing the system that produced it. Route well-scoped work to agents, track **human touches**, and make each harness or context improvement benefit the whole team through **shared registries** and clear ownership.

## Agent-Ready Context

The talk frames adoption across team, platform, and organization. Teams improve shared context and harnesses; platform owners curate reusable skills, evals, guardrails, and identities; leadership gives those owners a mandate rather than relying on scattered experiments.

Move retrospectives from patching agent-written code toward fixing the system that produced it. Route well-scoped work to agents, track **human touches**, and make each harness or context improvement benefit the whole team through **shared registries** and clear ownership.

The proposed dark factory is closer to a dim factory: autonomy should vary with feature risk, auditing, and verifier quality. The talk offers organizational patterns, not evidence that every team can reduce staffing or safely automate every workflow.

## Connected Context

Feed7 judgment across 545 accumulated Signals:

This raises coding-agent adoption from maintained team practice to an organization-wide ownership model: platform owners curate shared skills, evals, identities, and guardrails, while teams improve the production system after failures. It reinforces reusable context and evidence gates, but narrows autonomy through risk, verifier quality, auditing, and human-touch metrics rather than treating a dark factory as the default destination.

- [Agents, codebases, and teams — Aditya Khandelwal, Amazon AGI Lab](https://feed7.dev/p/agents-codebases-and-teams-aditya-khandelwal-amazon-agi-lab-1946kjc) — It extends that talk’s team-level repository operating system into platform and leadership responsibilities with shared registries and explicit ownership.
- [The Era of Compound Engineering — Kieran Klaassen, Every/Cora](https://feed7.dev/p/the-era-of-compound-engineering-kieran-klaassen-every-cora-053chml) — Its system-focused retrospectives operationalize compound engineering by turning corrections into shared harness and context improvements for later work.
- [How we set up our cloud agent environment](https://feed7.dev/p/cloud-agent-environment-1c839pq) — Cursor’s environment results identify a prerequisite for the proposed shared skills and registries: reliable, discoverable workflows may matter more than adding skills to fragile infrastructure.
- [Prototyping as Leadership: How a CTO Ships with AI Agents — Hursh Agrawal, The Browser Company](https://feed7.dev/p/prototyping-as-leadership-how-a-cto-ships-with-ai-agents-hursh-agrawal-t-1f1csz2) — Both reject uniform autonomy: bounded work and continued human review support the target’s dim-factory model for risk-sensitive adoption.

## Context Map

- Layer: agent
- Domains: coding
- Topics: harness-engineering, skills, enterprise

## Uncertainty

- The proposed dark factory is closer to a dim factory: autonomy should vary with feature risk, auditing, and verifier quality. The talk offers organizational patterns, not evidence that every team can reduce staffing or safely automate every workflow.

## Agent Instruction

Use this item as source-backed context. Do not invent claims beyond the linked source. If this item conflicts with another source, call out the conflict.