Hi, I'm Bryce. I'm an independent engineer (ex-eBay, nearly a decade) who designs and builds AI systems with Claude Code , MCP , and whatever else fits the problem. Daily since January 2025, across 260+ sessions: architecture, quality verification, prototype to production, and knowing when to scrap and rebuild. Solo builders, startups, established orgs. Bring me a project and I'll scope it honestly.
Now that AI writes the code, the hard part is everything around it.
What I Do
Prototype to production, end to end. Architecture with multi-agent coordination, quality gates, and self-improving memory. My flagship system runs an Etsy shop end-to-end with 34 specialized skills and three tiers of learned memory. I've evaluated and rejected two major orchestration frameworks against real requirements, and designed intent-based skill architectures that cut code 74% while increasing capability. I've also done a full architectural rewrite in 6 days when the original wasn't scaling, and scrapped 2 weeks of Chrome extension work after 8 PRs when a simpler approach was clearly better.
Problems found before they become yours. I've diagnosed undocumented bugs in Google's Gemini SDK, Anthropic's desktop software ( startup freeze , file upload failure , 6 VM failure modes behind one error ), and Electron's multi-monitor DPI handling. When standard troubleshooting fails, I trace root causes through log files, undocumented APIs, and system internals, then publish the fix and file the issue upstream.
Work you can trust. Every deliverable I publish goes through a 4-agent review pipeline (fact-checker, technical reviewer, editorial reviewer, visual QA). I've caught errors in my own drafts, retracted incorrect findings after publishing, and corrected community misinformation that was leading people to wrong fixes. When I'm wrong, I say so.
Just Published
Command: the design decisions behind an operator that must prove its fixes
New The key design decisions in Command, my autonomous Claude Code operator for 65 projects: a free deterministic watcher, per-repo config authored instead of loaded, eval gates before the engine existed, and a pre-registered success criterion committed to git before any fix runs.
Featured
Command
Active An autonomous Claude Code operator for 65 local projects. A free deterministic watcher senses the fleet, an approval-gated session acts, and every fix carries a success criterion committed to git before it runs.
The design decisions →ShopForge
Active Production AI agent that runs an Etsy shop end-to-end: from image generation to published listing to performance-based learning. 34 skills, three-tier memory; evolved from ShopSmith OS and live since early 2026.
active client engagement: Akaya knowledge management pilot (Oregon AI Accelerator '26 cohort), through Dec 2026
Latest posts
See all posts- Command: the design decisions behind an operator that must prove its fixesThe key design decisions in Command, my autonomous Claude Code operator: why the watcher is deterministic and free, why it authors per-repo config instead of loading it, why every fix carries a pre-registered success criterion, and what I want to see from it next.
- Parse the spec as claims: the CI drift guard behind an agent-readable design systemProse design specs drift the moment nobody's looking, and coding agents now read them as ground truth. The two-stage CI guard in chat-arch treats every palette-table row as a falsifiable claim: what it catches, the contrast bug that earned its own token, and the geometry drift it sailed right past.
- Adversarial Subagent Teams: A Retrospective71 invocations of the adversarial-subagent-team pattern in Claude Code and Cowork, verified against the cited PRs, commits, and arXiv references. Across six headline claims: three confirmed clean, one confirmed with caveats, one rejected on falsification, one not supported by the data.
New posts on Claude Code, MCP, AI tooling, and more. Delivered by email.
No spam. Unsubscribe anytime.
Work Experience
Full timeline- Jan2025 - NowIndependentSoftware Developer
Building production AI services, MCP protocol tooling, and e-commerce platforms as an independent developer. Primary development tool is Claude Code with AI co-authored commits across all projects.
Current projects include:
- Portland State University. Software engineering services on the Akaya knowledge management pilot — an AI-native platform that captures tacit team knowledge via voice and turns it into searchable playbooks, how-tos, and learning cards with citations. Akaya is a 2026 cohort company in the Oregon AI Accelerator. Through Dec 2026.
- Command. Autonomous Claude Code operator that runs 65 local projects through a free deterministic watcher and an approval-gated session, proposing per-repo config fixes as reviewable PRs and pre-registering each fix’s success criterion in git before it runs.
- ShopForge. Self-improving AI agent that runs an Etsy digital art print shop (ChateaucoreWalls) with three-tier memory, Bayesian confidence scoring, and skill evolution. Evolved from ShopSmith OS, which proved the domain across 86 commits and 37 merged PRs.
- Bryce Labs Toolkit. Claude Code plugin suite including spec-driven development, session history mining, and cross-platform skill management. Installable from marketplace.
- MCP API Bridge. Production-quality starter kit that turns any REST API into an MCP server for Claude, Cursor, and other AI assistants. 74 tests, released.
- Shopify TYOS Migrator. Automated migration of Shopify tracking pixels to the Customer Events API across 8 ad platforms with pixel parity validation.
Over 1,500 commits across 2025–2026 using a PR-based workflow with conventional commits, Claude Code co-authorship, and automated GPT-based PR review via chatgpt-codex-connector.
- Mar2015 - Dec2024eBaySoftware Engineer 2 → Member of Technical Staff 2
Nearly a decade at eBay, across Mobile Web, Charity, View Item, Android Native, and Ads. On Mobile Web, built much of the checkout page back when the whole app was a single codebase. Redesigned the eBay for Charity site to modern frontend standards, conversion climbed enough to clear our yearly team goals. On View Item, owned the Best Offer flow, among other surfaces. On the Ads team, modernized and consolidated a sprawl of applications into a more maintainable architecture. Promoted from Software Engineer 2 through Member of Technical Staff 2 across that span.
Production-scale systems work, code that has to stay up while you change it, is where I learned to ship without breaking things. That standard is what I now bring to the engagements on this page.
- Jun2012 - Feb2015Enterprise SoftwareSoftware Developer
Built enterprise payments and decision-analytics software. Developed systems for transaction processing, risk assessment, and data-driven decision support in regulated environments.
Recent projects
See all projects- Command ActiveAn autonomous Claude Code operator for 65 local projects. A free deterministic watcher senses the fleet, an approval-gated session acts, and every proposed fix carries a success criterion committed to git before it is allowed to run.
- MCP API Bridge ReleasedTurn any REST API into an MCP server for Claude, Cursor, and other AI assistants. Production-quality starter kit with 74 tests.
- ShopForge ActiveProduction AI agent operating a live Etsy shop with 34 specialized skills, three-tier memory, and graduated autonomy. Evolved from ShopSmith OS; live since early 2026.
- AI Research Pipeline ReleasedMulti-agent RAG research assistant with a 3-agent pipeline, intelligent model routing across o1, o1-mini, and sonar-pro, Chroma DB vector store, and Vectara factual consistency scoring.
- ShopSmith OS ArchivedLocal-first command center for managing and publishing digital products on Etsy with AI-powered content generation. Evolved into ShopForge.
- MCP Python SDK: PR #2041 Open PRContributed PR that exposed progress_callback in ServerSession methods, bringing Python SDK into parity with TypeScript. Awaiting maintainer review.
Let's Connect
For project inquiries, start at Work With Me . For anything else (speaking, collaboration, press), email me at hello@brycewatson.com or reach out on social.