Your first agentic workflow · Claude Code · Claude Cowork
In two weeks you'll know if it pays off.
We sit down with your people, take one real process out of your day-to-day and rebuild it as a governed agentic workflow — on your systems, under your rules. Measured against real cases, not a demo: you get proven numbers instead of an estimate.
Going live is your call afterwards — with the approval package already in hand. And every process we solve stays with you as a reusable building block: the principle behind Everything as a Skill.
2 weeks
1 process
12+
EU
The Problem
The gap isn't the idea. It's production.
Most AI engagements deliver a concept and leave the hard part — making it actually run inside your environment — to you.
✕ The pilot ran. The process is still manual.
A slide deck, a proof of concept, a roadmap — and nothing in production. Most AI initiatives fail not on the idea, but on the last mile — real data, real permissions, real approval from security.
✕ Knowledge stays in heads
One person figures out a brilliant agent workflow. It never becomes searchable, reviewable or reusable — and the learning curve resets with every project and every hire.
Why this is different
Not a consultant's deck. A process that runs.
I bring the setups, guardrails and workflows I run deep inside large, regulated organisations — and build them into your environment, with your people.
Built, not recommended
I don't hand over a concept for your team to implement. I sit in the same room and work in the same repository — the workflow runs against real cases before I leave, or the engagement isn't finished.
Everything as a Skill
Every process we automate is captured as a versioned, reviewed building block in your own library. The second process is faster than the first, and the tenth is faster still — because nothing starts from zero.
Compliant by design
GDPR-compliant model access with EU data residency (AWS Bedrock, Google Vertex AI), zero-retention options and EU AI Act awareness. Built with your DPO and CISO — not around them.
Infrastructure-deep
Devcontainers, sandboxing, permission guardrails, network egress control, audit logging. The layer most consultants skip — because they never had to pass an enterprise security review.
How we work together
Two weeks to a number you can prove
Every phase has a named result. If week 1 shows the process isn't worth automating yet, you'll hear that — and we pick a better one.
Pick & map the process
In the room: process owners, the business side, engineering
We don't start with the tool. We start with the process that hurts — and we make its current cost measurable, so "better" isn't a matter of opinion later.
- ▸Process interviews with the people who run it daily
- ▸Candidate scoring by effort, frequency and risk
- ▸Baseline measurement — cycle time, error rate, cost
- ▸Clear split — what the agent does, what humans decide
- ▸Data access, permissions and approvals clarified up front
Outcome: One selected process, documented, with a measured baseline — and agreement on what "better" concretely means.
Build & secure the workflow
In the room: engineering, security & data protection
We build the workflow on your real data and systems — with the guardrails your security team needs, from the first commit rather than bolted on afterwards.
- ▸Agentic workflow on your real data and systems
- ▸MCP connections to Jira, GitLab, databases, line-of-business systems
- ▸Guardrails — what the agent may do, approval gates where needed
- ▸Devcontainer sandboxes instead of laptop-wide access
- ▸Audit trail and cost controls from day one
- ▸Tested against real cases, not sample data
Outcome: The workflow runs against real cases — with guardrails, audit trail and an after-value you can hold against week 1.
Hand over & make it stick
In the room: your team, your future champions
The engagement ends with your team driving, not with a dependency on me. What we built becomes a template for the next process.
- ▸Handover — your team drives, I watch
- ▸The process becomes a versioned skill in your library
- ▸Reusable pieces become standards (MCP, guardrails, conventions)
- ▸Playbook for the next process — the second one is faster
- ▸4 weeks of async support
Outcome: Your team builds the next process without me — and the first one stays with you as a reusable building block.
Processes we've rebuilt
Processes, not demos
A selection of what we build together — engineering and business processes alike, drawn from real engagements.
From ticket to release candidate
Changelogs, release notes, migration checks and pipeline steps drafted and verified by agents — the release stops being an evening job for one person who knows the ritual.
Diagnosis runs before the human wakes up
The agent gathers logs, metrics and recent changes and files a first hypothesis with the alert. Postmortem findings become executable skills instead of wiki pages nobody reads at 3 a.m.
Audit evidence on autopilot
Compliance evidence generated and versioned continuously by agents, instead of hand-assembled in panic in the weeks before every audit.
Framework upgrade across dozens of repos
Upgrades and API migrations fanned out over the whole estate, each change verified by its own checks — weeks of mechanical work compressed into days.
Answering tenders instead of copy-pasting them
Claude Cowork drafts responses from your past bids, product docs and requirement lists with governed data access — the expert reviews and decides instead of assembling.
Cost per case, not per surprise
Lean skills instead of bloated prompts, company-wide MCP standards that cap file and output context, and subagents that route routine work to smaller models. Cost per task drops by an order of magnitude — and stays visible per team and process.
Why it survives the security review
This is where pilots turn into production
An agentic workflow that can't be approved never ships. The infrastructure underneath is what turns a working demo into a process your organisation is allowed to run.
EU model gateway
Central, GDPR-compliant access to frontier models via AWS Bedrock or Google Vertex AI — EU data residency, zero retention, no training on your data. Egress control and DLP keep your crown jewels out of prompts and logs.
Devcontainer sandboxes
Agents run in disposable, reproducible containers with scoped credentials — not on developer laptops with domain-wide access.
Guardrails & policies
Permission allowlists, hooks and approval gates define what agents may do — deterministically, not by trusting the model. SSO and role-based access for every human and every agent.
Observability & audit
OpenTelemetry-based usage logging, audit trails and dashboards for adoption, quality and cost.
>_ Designed so your CISO and DPO say yes — documented so your auditors do too.
Engagement models
Start with one process
Clear scope, fixed outcomes — priced for organisations that expect a running process, not a workshop certificate.
Workflow Sprint
Fixed scope, fixed outcome — one real process, proven against real cases, with your people building alongside me.
- ✓Process selection with a measured baseline
- ✓One agentic workflow, running on your systems
- ✓Guardrails, sandbox and audit trail to your standards
- ✓Before/after numbers you can take to your leadership
- ✓Approval package for going live — architecture, data flows, permissions, audit concept
- ✓4 weeks of async support afterwards
For: one team with one process that hurts
Process Program
Several processes of one team — plus the secured platform and the skill library that make the next ones fast.
- ✓3–5 processes rebuilt as agentic workflows
- ✓GDPR-compliant platform per the reference architecture
- ✓Your skill library — seeded, versioned, owned by you
- ✓Cost architecture — model routing, token budgets, caching
- ✓Enablement of your team along the way, not as a separate course
For: teams of 5–15, incl. platform & security
Embedded Partner
A standing partnership that keeps turning processes into workflows — and keeps the growing library governed.
- ✓Quarterly process delivery with your teams
- ✓Skill library governance and curation
- ✓Champions program & internal enablement
- ✓Adoption, quality and cost reporting to leadership
For: organisations scaling beyond the first team
Solo professional or small team? Ask for the compact one-day intensive — one small process plus a secure single-seat setup.
Next step
Bring your hardest process
Book a 30-minute call and bring one real process that hurts. I'll show you, concretely, how it looks as a governed agentic workflow — with the guardrails your security team will demand. If I can't, you've lost half an hour, not a budget.