Unreliable AI-generated changes
Changes that look plausible can miss requirements or introduce regressions. Establish clear acceptance criteria and checks before code reaches production.
Clear. Hype-free. Experience.
Make agentic coding a reliable way to deliver software
We help SaaS teams improve delivery predictability, software quality and control of AI delivery costs.
Arrange a 20-minute conversation
About
Mick Clarke
Founder of Crepuscular
I help teams turn AI adoption into reliable software delivery, drawing on hands-on work in SaaS and government software. My work spans adoption strategy, architecture, agentic workflows and the tools that make quality visible.
LinkedInConsulting
Make AI work across the delivery workflow.
Agentic coding changes how work is created, reviewed and shipped. We help your team find and fix the points where that process breaks down.
Let’s work through itExcessive review effort
When review takes longer than generation, speed disappears. Improve context, change boundaries and evidence so reviewers can make confident decisions.
Unclear costs
Token spend is only part of the bill. Make retries, review time and rework visible alongside tool costs to understand what delivery actually costs.
Fragmented delivery workflows
Disconnected prompts, tools and hand-offs make work hard to follow. Connect intent, implementation, checks and release into a workflow the team can repeat.
A place to start
AI Delivery Review
A focused review of one team and one delivery workflow, producing prioritised improvements and a scoped pilot recommendation.
01 / Understand
One team. One workflow.
Look at how an AI-assisted change moves from intent to release, and where reliability, review effort and costs become difficult to manage.
02 / Prioritise
Improvements in order.
Identify what to change first, with the reasons and trade-offs made clear.
03 / Test
A scoped pilot recommendation.
Define a practical next step and the outcomes that would show whether it is working.
We agree the scope and fee before work begins.
Recent work
Practical work. Credible evidence.
Government software provider
From adoption to an AI Factory.
I led organisation-wide AI adoption, from none to >90% in three months, then built an AI Factory to optimise AI engineering delivery combining requirements, modular architecture and quality tooling. A four-person team delivered an externally assured revenue-management product in under six months.
New Zealand SaaS team
A measured start with agents.
I helped an AI task force shape a bounded pilot, with architectural advice and QA tooling exemplars. In the first week, fourteen runs produced five CI-confirmed bug-reproduction PRs; nine declined with named blockers. All decisions matched human judgement. Follow-up work automated the process of bug reproduction and recommended resolution.
Crepuscular
The foundations
behind better work.
We’re building a modular approach to agentic engineering. Each part has a clear purpose. Together, they free you to focus on your business
01
Platform
A shared foundation for agentic work. Clear boundaries, repeatable operations and visibility from the start.
Talk about platform
02
Agents
Give agents the context, tools and limits they need. Keep people responsible for intent, decisions and exceptions.
Talk about agents
03
Ontology
Clear tracing of intent to evidence. Make requirements, relationships and the history of work explicit, token efficient and searchable. Out of the box auditable history of action.
Talk about ontologyOur approach: define the work, bound it, automate it, prove it and make it visible.
Speaking & outreach
The AI Factory.
My KiwiSaaS talks explore the AI Factory: designing the whole software delivery process around agents, with clear intent, repeatable work and independent evidence of quality.
01Machining as a job
What can software learn from manufacturing? A conversation about standard work, quality built into the process and the craft of making things reliably.
02Why the sidecar doesn’t work
Adding AI to the same engineering workflows can make your current constraints crippling bottlenecks. Explore what changes when the whole process is designed around agents.
Contact
Arrange a 20-minute conversation
Tell us about your team, the workflow you want to improve and what is getting in the way. We’ll use the conversation to see whether an AI Delivery Review is a useful next step.