Chris Chabot

Chris Chabot, Member of Technical Staff, building agent platforms

I design and build the systems that take agents from demonstration to infrastructure: runtime and orchestration, SDK contracts, memory and permissioning, and the evaluation and verification that decide what ships.

Loose strokes of rust, dark green, plum and ochre cross scraped chalk openings.

Selected work

  1. Catching what coding agents leave out in a large port

    A production coding agent of about 340,000 lines of Python, built by my colleagues, moved to TypeScript with one requirement: byte-identical behaviour in everything not specific to the new platform. I designed the method and directed the agents: a formal model, a checklist generated from the syntax tree, clean-context review, then replay.

    designed, directed, reviewed, verified

  2. Keeping release testing finite

    Release qualification for an agent-native cloud platform, designed around what a customer actually does. Evidence comes in layers, from a certified core to real browser journeys through sample apps, and a result is reused only where it still applies.

    designed, verified, operated

  1. Keeping multi-language SDKs aligned with conformance fixtures

    Five languages have no single implementation to share, so the contract had to be data: fixtures for wire shapes, error codes and verified identity that the SDKs' own tests read.

    designed, led

  2. Checking that an agent experiment actually ran your change

    Before asking whether a change helped an agent, I want to know whether the comparison actually contains the change. The tooling here treats each run's configuration as a claim to be checked, and keeps corrections beside the original record rather than over it.

    designed, built

  3. Status models for long-running agent projects

    What changes when a coding conversation becomes a project that keeps running while you are elsewhere, and what a coordinator and its workers need in order to stay useful when there are a hundred of them.

    designed, directed

  4. Verifying agent memory from capture to delivery

    Memory has to be isolated by tenant, keep the current fact in front, and stay out of the way of the next reply.

    designed, built

  5. Binding human approvals to a single tool call

    An approval should mean one tool call, with these exact inputs, in this thread and turn, under this version of the policy, once.

    designed, built

Public code

All public projects

Earlier products, open source and standards

  • Apache Shindig for PHP

    Chief author of the PHP implementation of the OpenSocial container; PMC member and committer.

    built, from 2008

  • OpenSocial

    A common API for social applications across participating networks.

    co-authored, contributed, from 2007

  • Portable Contacts

    A simple XML and JSON protocol for address books and friends lists over HTTP.

    co-authored, 2008

  • OAuth 2.0

    The framework for granting applications limited, scoped access to an HTTP service.

    contributed, 2012

  • OpenID Connect

    An interoperable answer to "who signed in?" on top of OAuth 2.0.

    contributed, 2014

  • Google+

    Founding member of the project; led the platform's developer launch and API ecosystem.

    contributed, led, from 2010

All earlier work

Latest writing

  1. Reporting compatibility by capability instead of a single percentage

    What one compatibility percentage hides about a partial browser engine, shown by comparing a Rust engine built for AI agents with Chrome across 1,024 sites and reporting each capability separately.

    Agent systemsEvaluationPlatforms

  2. What to simulate in an agent browser and what to leave to the host

    A headless browser for agents that honours page events exactly but returns form submissions as data, leaving network, transport and session state to the host.

    BrowsersAgent systemsRuntimeArchitecture

  3. Verifying large data imports without running them again

    A large import passes through a source fixture, uploaded objects, a published version and a reader view, each with its own count, and can be verified read-only without importing again.

    DataImportsVerificationFerrite

  4. Preemptive timeouts for untrusted JavaScript

    A timer on the same event loop cannot stop JavaScript that never yields. Enforcing the deadline from a separate process, and what a process kill then requires.

    Agent systemsPlatformsSecurity

  5. What each layer of a browser test suite actually proves

    What screenshot comparisons, end-to-end journeys, store readback and honest reports each prove about a browser app, and why Undo needs an invariant rather than one engine's habits.

    VerificationBrowsersReact

All technical writing

Latest essay

The boat at night

June 2026

All essays

Dark ink blue and black passes drag diagonally from upper right, opening a crooked chalk strip with an ochre smear and silver-grey scumble.