# chen-fioca-openai-future-proof-coding-agents-2025-11-23

## Veille

Building durable coding agents that survive model cycles: Harness architecture, Codex SDK, and terminal Computer Use — OpenAI's experience report

## Titre Article

Future-Proof Coding Agents: Building Reliable Systems That Outlast Model Cycles

## Date

2025-11-23

## URL

https://www.youtube.com/live/cMSprbJ95jg?si=4HnxK8w1ELvSr4tz&t=7320

## Keywords

OpenAI, Codex, Coding Agents, Harness, SDK, Computer Use, Agentic Patterns, Model Abstraction, Infrastructure

## Authors

Bill Chen (Product Manager, OpenAI), Brian Fioca (Engineering, OpenAI)

## Ton

**Profile:** Technical-Architectural | Systems Engineering | Pragmatic | Expert

The tone is that of engineers and product leads sharing lessons learned "in the trenches." The style is didactic yet technical, focusing on the anatomy of agents (Harness, Model, UI) and integration challenges. The approach is one of infrastructure problem-solving: how to abstract away the complexity of changing models in order to build durable tools. The discourse targets developers of tools and agents, using precise vocabulary (compaction, context window, sandboxing, MCP).

## Pense-betes

- **Anatomy of an agent**: User Interface + Model + Harness (the "harness").
- **The Harness is critical**: it is the interface layer that manages complexity (tools, prompts, memory, security).
- **Harness challenges**:
- Adapting to new models (habits vs. intelligence).
- Managing latency and UX "while the model is thinking".
- Managing the context window and compaction.
- Security (sandboxing, permissions).
- **Codex as Harness**: OpenAI offers Codex not only as a model but as a "batteries included" agent/harness (tool management, CLI, sandboxing).
- **Computer Use for the Terminal**: Codex acts as an agent capable of using the computer via the terminal, unifying coding tasks and system administration.
- **Emerging patterns**:
- The Harness as a new abstraction layer (avoiding rewriting prompt engineering for each model).
- SDK Codex for integrating the agent into other products (e.g., GitHub, Cursor).
- Agents building their own tools (dynamic MCP connectors).

## RésuméDe400mots

Bill Chen and Brian Fioca of OpenAI's Applied AI team present a methodology for building durable "Coding Agents" capable of surviving rapid model evolution cycles. They break down the anatomy of an agent into three parts: the User Interface, the Model (the raw intelligence), and the "Harness" (the integration infrastructure).

The core of their talk focuses on the underestimated complexity of the **Harness**. Building a robust harness is difficult because it must handle adaptation to each model's specific "habits" (prompt engineering), latency management, context compaction, tool integration (MCP), and above all, security (sandboxing). They point out that models have "personalities" or training biases (such as the tendency to want to read everything before acting) that the harness must manage in order to be effective.

To address this problem, OpenAI positions **Codex** not merely as a model but as an integrated "agent + harness" solution. Codex (the agent) natively handles the complexity of interacting with the system: running terminal commands, editing files, managing memory, and creating tools on the fly. This makes it a "Computer Use Agent" for the terminal, capable of carrying out tasks ranging from pure coding to system administration or data analysis (CSV files, folder organization).

They identify emerging patterns for agent developers:
1.  **Harness as an abstraction layer**: Using a robust harness (such as the SDK Codex) allows teams to focus on product differentiation rather than on maintaining low-level infrastructure (prompting, tool calling).
2.  **Agents within Agents**: Integrating Codex via SDK as a "sub-agent" capable of executing complex technical tasks within a larger application.
3.  **Dynamic tool creation**: The ability of agents to write their own connectors (MCP) to interface with APIs or systems without prior integration.

In conclusion, they encourage developers not to "reinvent the wheel" of the harness each time, but instead to rely on existing infrastructure to build richer, more stable user experiences.

## GrapheDeConnaissance

- Bill Chen —recommande→ architecture agents codage (CONCEPT, 0.92)
- Brian Fioca —recommande→ architecture agents codage (CONCEPT, 0.92)
- OpenAI —emploie→ Bill Chen (PERSONNE, 0.95)
- OpenAI —emploie→ Brian Fioca (PERSONNE, 0.95)
- OpenAI —a_créé→ Codex (TECHNOLOGIE, 0.97)
- Codex —utilise→ Harness (CONCEPT, 0.92)
- Harness —résout→ complexité intégration modèles (CONCEPT, 0.9)
- Codex —utilise→ Computer Use terminal (METHODOLOGIE, 0.88)
- agents IA —a_créé→ outils dynamiquement (MCP) (TECHNOLOGIE, 0.85)
- SDK Codex —permet→ intégration sous-agent (CONCEPT, 0.88)
- Harness —réduit→ dépendance aux cycles d'évolution des modèles (CONCEPT, 0.9)

---
Canonical: https://www.thekb.eu/en/fiches/chen-fioca-openai-future-proof-coding-agents-2025-11-23/