Stop repeating the investigation
Make relevant findings and decision history available when an agent needs them. Keep sources and applicability alongside the advice.
Engineering for teams working with coding agents
Make your coding agents
Make your coding agents
work better together.
Shared knowledge. Useful context. Changes you can trust.
Your agents can write code. I help them find the knowledge, use the tools and run the checks they need to deliver work your team can build on.
Work directly with David · 20+ years building software · 8 years at Zendesk
Every useful discovery should become a head start.
Give agents the right connections
Connect your APIs and engineering tools through usable integrations, with clear access boundaries and ownership.
Know whether the work is done
Define meaningful acceptance checks and capture evidence that a change works before it reaches review.
Improve the workflow your team already uses.
Shared engineering knowledge
Reusable findings, project context and decision records that survive a session handoff.
Working integrations
Repository and MCP changes that connect your agents to the tools and knowledge they need.
Evidence your team can inspect
Acceptance checks, verification results and a walkthrough of the delivered work.
Start with one representative workflow.
Discuss the friction
Show where your agents repeat work, lose context or require too much supervision. We agree what improvement would mean.
Assess the workflow
Trace a real task from context gathering to review. Receive prioritized findings and a concrete implementation proposal.
Implement and evaluate
Install the agreed improvements, exercise the workflow and compare results with the baseline. Your team owns review and release.
Scoped to your team.
Agent workflow assessment
A focused review of context, shared knowledge, integrations and verification. You receive prioritized findings, a baseline and a practical implementation scope.
Implementation and handoff
Hands-on changes in your repositories and tools, with agreed acceptance checks, verification evidence and a team walkthrough.
Fixed fee, timeline and deliverables agreed in writing before work begins. Assessment and implementation can be commissioned separately.
Found me through Ajent? Bring the knowledge-sharing or coordination problem behind your agent’s question. We can work with your existing tools.
The tools behind my practice
I build and use tools across the agent workflow. Explore the code to assess my engineering approach; these projects are evidence of the work, not claims of client results.
Ajent ↗
A place for coding agents to share findings and find relevant knowledge from other agents. It explores how useful discoveries can travel beyond the session that produced them.
Kazi ↗
Express a goal as machine-checkable conditions and give a coding agent a concrete finish line. Relevant when a successful agent run still leaves you unsure whether the task works.
Dira ↗
A git-native ledger of decisions and the alternatives they rejected. Keep the reasoning behind a change available when another agent or developer picks up the work.
Serenity ↗
A personal memory and direction system being built around a git-backed record. It explores how agents can share durable context while keeping consequential changes under human review.
Zerfoo ↗
A machine-learning inference and training framework in Go. This is the infrastructure side of my work: model execution, performance, and the systems beneath agent applications.
Mint ↗
Generate MCP servers from OpenAPI specifications. Relevant when an agent needs a usable connection to your APIs and internal systems.
Sire ↗
My platform for durable agent workflows and human approval gates. It is also where I am applying these ideas to my own operations.
Found me through one of these projects? Tell me which one and what you are trying to do. I can help evaluate fit or implement it in your workflow.
How much agent work I actually run
Token throughput from the Claude Code sessions I run daily, pulled automatically from local session logs across my machines. Personal activity only. Token volume is not evidence of quality, savings or client outcomes.
—
tokens processed · last 24 hours
—
tokens processed · last 7 days
—
tokens processed · last 14 days
—
machines reporting
I’m David’s AI assistant. What are you using coding agents for, and what still takes too much of your time?
Before we work together
Do we need to adopt Ajent?
No. Ajent is part of my work on shared engineering knowledge. The engagement starts with your workflow and existing tools; a product subscription is not required.
What does an engagement cost?
We agree on a fixed fee after defining the scope, deliverables and acceptance checks. The discovery call is free and creates no commitment.
How do we evaluate success?
Choose a representative task and record the current friction. Agree on observable checks for the delivered improvements; no generic savings guarantee replaces that evidence.
Where could your agents work better?
Tell me which agents you use, where work repeats or needs too much supervision, and what you want to improve.