Owain Williams Agentic engineering
PrincipalSoftwareEngineer.
I build codebases that AI and humans alike can navigate with ease.
Vibe coding makes a fine prototype, not a business. Bad code has never been more expensive, and a big ball of mud has never been easier to build. Deep systems design is what turns agents into build velocity that wasn't possible five years ago.
- TeamHuman direction
- AgentsAgent orchestration
- DataData foundations
An exploded diagram of three stacked layers: a team layer giving human direction on top, an agent orchestration layer in the middle, and a data foundations layer underneath, connected by orange traces.
const system = {
team: buildTeams(),
agents: orchestrate(),
data: buildFoundations()
};Engineering, in practice.
Case study directionsGrowing a product engineering group without losing the clarity of a small team.
- Context
- A team had doubled in a year. Ownership was blurred, reviews queued for days and nobody could say who decided what.
- Decision
- Re-drew the team boundaries around the system's seams, wrote down decision rights and made architecture reviews a weekly ritual instead of a gate.
- Trade-offs
- Slower in the first month while the new boundaries settled. In return, far fewer hand-offs and a team that could explain its own system.
Leadership is part of the architecture.
Think clearly. Build deliberately.
Team / Human direction
People set intent, make the trade-offs and own the outcome. Structure the team the way you would structure the system.
Agents / Agent orchestration
Agents do narrow, observable work between clear review points. They are designed, tested and owned like any other service.
Data / Data foundations
Owned, versioned data that both people and agents can trust. Everything above depends on getting this layer right.
Expertise
- Team leadership
- AI-first engineering
- Agent orchestration
- Data-first systems
Engineering Notes
Systems // Teams // Agents // Data- Vibe coding is not a business plan.Vibe coding is brilliant for a prototype. Run a business on it and you own a big ball of mud that nobody, human or AI, can change safely.
- Agents need architecture.Agentic engineering is a skill. Agents move fastest in codebases with clear seams, contracts and names, so the architecture comes first.
- Engineering decisionsWritten decisions are context for people and agents alike. A short template: context, decision, trade-offs. Nothing more.
About
Working historyI'm a principal engineer who treats the team, the agents and the data as one system, because that's how they behave in production.
Most of my work sits where people, process and architecture meet: setting technical direction, growing engineers into leaders, and bringing AI into everyday engineering in a way teams can trust. I like plain language, written decisions and systems that are boring to run.
I've spent my career helping ambitious small and medium businesses ship better software, faster, without burning out the people who build it.
- 2026 — Now
Principal Software Engineer
Cavis Marketing Ltd
Drive Engineering initiatives across teams, AI-first practice and leadership.
- 2024 — 2026
Software Engineer
Ultramed
Medical software. Robust, scaleable, compliant
- 2022 — 2024
Technical Project Manager
Energise
Build a development team from the ground up. Migrate legacy systems to modern architectures.
- 2013 — 2022
Software Engineer
Merlin Systems Ltd
Write robust, testable, maintainable code. Install and manage on-premises Linux servers and infrastructure
- Drawing
- The Engineer's Working Atlas
- Sheet
- 02 / 02
- Rev
- A — first layout
Need a principal engineer's help?
I'm happily employed and not looking for a new role. I am open to consulting work, as long as it's legal and doesn't conflict with my current employment. Tell me about the team, the system and what needs to move.