llms.txt
Compare with other sites' llms.txt files →# The Davenporter — Jason Davenport (@davenportjw)
> Learnings about software, technology, agentic systems, and occasionally home renovation. Exploring agent-first source control (cosm), trajectory-to-lineage observability, Spanner-backed agent memory consolidation (Electric Sheep), hybrid local/cloud LLM routing, and real-time multiplayer browser simulations on Google Cloud Run.
## Summary
I'm Jason Davenport. I write about what happens when agents do real engineering work: source control, memory, telemetry, and the human side of AI. Here are my essays, the code behind them, and a game or two.
## Verified Identity & Endpoints
- Curator: Jason Davenport (@davenportjw) — AI & Distributed Systems Architect
- Canonical Site: https://davenporter.com
- Substack Publication: https://davenporter.substack.com (RSS: https://davenporter.substack.com/feed)
- GitHub Profile: https://github.com/davenportjw
- LinkedIn Profile: https://www.linkedin.com/in/davenportjw
- X / Twitter: https://x.com/davenportjw
- Complete Corpus: https://davenporter.com/llms-full.txt
## Pages
- [Home](https://davenporter.com/): Latest updates, essays, code, and games.
- [About](https://davenporter.com/about): Who I am, my career arc, what I build, and life in Colorado.
- [Essays](https://davenporter.com/writings): All 23 Substack essays, each with a short summary.
- [Builds](https://davenporter.com/builds): Freight Floor (a multiplayer forklift game) and open-source tools for AI agents: cosm / topocosm, SmartRouter, and Electric Sheep.
- [Code](https://davenporter.com/code): 6 code projects, each linked to the essay that explains it.
- [Now](https://davenporter.com/changelog): Recent essays, GitHub updates, and Cloud Run deploys.
## Production Code Blueprints (@davenportjw)
- [ast-graph](https://github.com/davenportjw/agent-skills/tree/main/plugins/ast-graph): ast-graph: a code graph so coding agents read less — ast-graph is a plugin for Antigravity and Claude Code. It parses your repo into a local SQLite graph of symbols, line ranges, calls, field reads and writes, routes, and tests. The agent queries the graph through MCP tools before it opens a file, so it reads the lines it needs and runs the tests that cover the change.
- [cosm / topocosm](https://github.com/davenportjw/cosm): cosm and topocosm: source control for AI coding agents — cosm stores your code as a graph of functions, classes, and the calls between them instead of as files. Before editing, an agent asks what else the change could break, takes a short lock on just that piece, and cosm writes normal files back out so your usual build and test tools keep working. topocosm is the shared server that coordinates many agents working in one codebase.
- [SmartRouter](https://github.com/davenportjw/smartrouter): SmartRouter: send each prompt to the cheapest model that can handle it — SmartRouter sits in front of your models and decides, one prompt at a time, whether to run it locally on Gemma 4 or send it to Gemini on Google Cloud. It scores how complex the request is, checks the token budget and any data-residency rule, and picks a route that fits your latency and cost limits. It records why it chose each route.
- [Electric Sheep](https://github.com/davenportjw/agentic-memory): Electric Sheep: long-term memory for AI agents — Electric Sheep gives coding agents memory that lasts across sessions. Between sessions it runs a 'dream cycle': it groups similar events from past sessions and promotes a pattern to a lasting belief only after it has held up across several of them. Beliefs live in Cloud Spanner and agents read them through an MCP server.
- [agent-telemetry](https://github.com/davenportjw/agent-telemetry): From Trajectories to Lineage: OpenTelemetry Span & Commit Provenance — Bridges the gap between ephemeral LLM reasoning traces and permanent repository commit history by binding OpenTelemetry span IDs, tool-call trajectories, and token budgets directly to AST/Git mutations.
- [nano-browser-agent](https://github.com/davenportjw/the-davenporter): Browser-Native Agent: On-Device Gemini Nano & WebMCP — Browser-native multi-turn docent architecture running on-device via Chrome's LanguageModel and Summarizer APIs. Uses document.modelContext (WebMCP) to expose grounded client-side tools and falls back dynamically to Google Cloud Gemini platform (gemini-3.8-flash) via Application Default Credentials.
## Essays (23)
- [Your coding agent is reading too much code](https://davenporter.com/writings/ast-graph-stop-reading-every-file) (Nov 06, 2026, 5 min): I built an AST plugin so agents jump to the exact lines, callers, and tests they need.
- [WIBTM: Demos should be fun, why I built FreightFloor (and how the AI works)](https://davenporter.com/writings/freightfloor-demos-should-be-fun) (Oct 27, 2026, 4 min): You aren't going to just look at a log. Here are the top three takeaways from building a multi-agent warehouse game in 5 calendar days.
- [100% of your developers should be building things to improve the workflow. Are they?](https://davenporter.com/writings/workflow-ownership-fix-the-machine) (Oct 21, 2026, 4 min): Developers need to fix the machine, not the execution output.
- [WIBTM: What happens when you let an agent design a new git system?](https://davenporter.com/writings/wibtm-what-happens-when-you-let-an) (Sep 24, 2026, 5 min): Why git and AST’s have some relationship, and what’s next when it comes to systems that are agent-first.
- [The value of being bored in the AI and attention economy](https://davenporter.com/writings/the-value-of-being-bored-in-the-ai) (Sep 16, 2026, 3 min): And how it can keep us in the high value areas of AI engineering.
- [What happens AI burns down the backlog?](https://davenporter.com/writings/what-happens-ai-burns-down-the-backlog) (Sep 03, 2026, 5 min): AI is still a business transformation problem, and its out-kicking the product coverage.
- [From trajectories to lineage: building better agent telemetry](https://davenporter.com/writings/from-trajectories-to-lineage-building) (Aug 27, 2026, 3 min): A conceptual map of how to establish provenance for agent decisions using the existing tools we have today
- [Is the future of learning ‘vibe’ training?](https://davenporter.com/writings/is-the-future-of-learning-vibe-training) (Aug 25, 2026, 9 min): Unpacking how we teach non-determinism in a 30 second attention span economy.
- [WIBTM: Add telemetry and observability to your AI agents](https://davenporter.com/writings/add-telemetry-and-observability) (Aug 18, 2026, 5 min): Telemetry and observability give you the means to start capturing data that make your agent systems long running, and improve their capabilities over time.
- [WIBTM: Give an agent access to memories created with AI dreams](https://davenporter.com/writings/give-an-agent-access-to-memories) (Aug 12, 2026, 4 min): I gave this talk at AI4 on how to create a memory system for your agents to dream, sharing here in the long form.
- [The State of AI in the SDLC: A Roadmap for Scaling](https://davenporter.com/writings/ai-sdlc-scaling-framework) (Jul 01, 2026, 6 min): AI innovation is messy when you're trying to ship. Here are some practical things I've found to integrate AI into your software development lifecycle, from your first customer to your thousandth.
- [Moving Past "Demos Over Memos": The Future of AI in Enterprise Product Development](https://davenporter.com/writings/emos-over-memos-enterprise-ai) (Jun 24, 2026, 4 min): Vibe coding and pure prototypes are creating cognitive overload. Here is how enterprise engineering teams must restructure for long-term AI success.
- [How to Track AI Agent Lineage and Manage State in Code Repositories](https://davenporter.com/writings/how-to-track-ai-agent-lineage-and) (Jun 16, 2026, 6 min): Moving beyond clean git commits to knowledge systems for agentic development.
- [WIBTD/W/M: A fool's folly with local AI models](https://davenporter.com/writings/wibtdwm-a-fools-folly-with-local) (Jun 11, 2026, 6 min): Failures, learnings, and hopefully good next steps to being realistic about developing AI models and methods
- [With AI, what's old is new in telemetry again](https://davenporter.com/writings/with-ai-whats-old-is-new-in-telemetry) (Jun 09, 2026, 4 min): Why tracing is critical to improving your AI experiences
- [WIBTD/W/M: Routing AI responses between local runners](https://davenporter.com/writings/wibtdml-routing-ai-responses-between) (Jun 04, 2026, 5 min): Extending smartrouter to support calling local AI models that are a part of a pool of workers, learnings and tribulations, and things I'm planning to do next
- [What's worth learning in an AI era?](https://davenporter.com/writings/whats-worth-learning-in-an-ai-era) (Jun 02, 2026, 4 min): How AI has changed the question from how to learn quickly to if learning will be useful.
- [What I'm building this week: Route your Gemini API calls on your terms](https://davenporter.com/writings/what-im-building-this-week-route) (May 28, 2026, 3 min): A little router I'm building and improving to use different Gemini (and soon other) models.
- [Can you 'learn' to use AI without being all in?](https://davenporter.com/writings/can-you-learn-to-use-ai-without-being) (May 26, 2026, 8 min): Explorations about successful learning patterns for AI and how I'm using it currently
- [What I'm building this week(/month/quarter)](https://davenporter.com/writings/what-im-building-this-weekmonthquarter) (May 21, 2026, 2 min): This week: How agent skills are improving my session workflows
- [AI and the New Business Moat: Defending Your Competitive Advantage](https://davenporter.com/writings/ai-and-the-new-business-moat-defending) (May 19, 2026, 5 min): AI requires you to create another moat of advantage. Here are my thoughts about how to do it.
- ['Dark data' is your new most valuable data for AI](https://davenporter.com/writings/dark-data-is-your-new-most-valuable) (May 13, 2026, 3 min): And why big data is making a comeback to bring data out of the shadows that your agents need in order to be successfully created and trained.
- [Maximizing Developer Value: Using AI Across the Entire Code Lifecycle](https://davenporter.com/writings/maximizing-developer-value-using) (Jul 10, 2025, 6 min): There’s now a wealth of information and tutorials floating around about how you can go from 0 to 1 with AI, or even vibe code personalized apps or prototypes?