Leading Teams That Bring Agentic AI into Production
25 years in software, the last years leading delivery teams of up to 35 people. My teams build agentic software for large enterprises. I care about what happens after the demo: delivery that is deterministic, auditable, and ready for real users.
Open source: aSPARK – a methodology and toolchain that brings determinism and auditability to software delivery in mixed teams of people and agents. MIT.
At a glance
- 25+ years of shipping software, coding since 1996
- up to 35 people in the delivery teams I have led
- 3 industries automotive, manufacturing, logistics
- 5 open source products in the aSPARK family
What You Will Find Here
The focus is actionable knowledge over hype: concise guides, deep dives, honest retrospectives, and templates you can apply directly in real projects.
-
01
Agents in Production
Real-world notes on planning, prompting, tool usage, and quality controls in agentic workflows.
-
02
Leading AI Delivery
Team setup, roles, and what it takes to get from pilot to production at enterprise clients.
-
03
Governance and Auditability
Determinism, traceability, and clear roles for teams where people and agents ship together.
aSPARK: determinism for teams of people and agents
Agents make teams faster. They also make it harder to say who decided what, which rule applied, and whether the result can be reproduced. aSPARK is my open answer to that.
- Gates: every feature passes five phases, and only a person opens the next one
- Traceability: stories, code, tests, and releases linked in a local knowledge graph
- Policy as code: engineering standards that are checked, not just written down
Latest Posts
-
Why the Forward Deployed Engineer role needs to be understood now
What I'm seeing with industrial and supplier clients right now, why agentic software development needs explaining, and what a Forward Deployed Engineer can do in that kind of change.
- agentic-workflows
- best-practices
-
My agent had no stop condition
Why a feature loop doesn't fit a migration, what I was missing for long mechanical work, and what an agent that halts on its own looks like.
- agents
- claude-code
- agentic-workflows
-
From a 47-Minute Screen Recording to a Narrated Demo — Without Opening a Video Editor
How a raw recording of one aSPARK session became a captioned, branded, narrated 100-second demo the same day — planned from a hash ledger instead of a timeline, rendered with what ships with macOS, and voiced with ElevenLabs.
- claude-code
- agents
- agentic-workflows