Consulting · Services · Research
Landing Agentic AIAgents, Minus the Theatre
I help corporations take agentic AI from demonstration to daily operation: which processes to start with, how to build, connect and govern the agents, and how to measure what they change.Everyone has an agent demo. I help you get one that turns up for work, follows the rules and pays for itself.
Where agents stall
Why most agents never leave the demo.How a good demo dies.
It was proven on someone else’s data.The demo ran on the vendor’s data.
An agent that worked in a sandbox meets your real documents, exceptions and permissions for the first time in production.
Nobody decided what it may decide.It has admin rights and no job description.
Without written limits on what the agent can approve, change or send, every incident becomes a debate about whose fault it was.
It cannot reach the systems of record.It can chat. It can’t touch the ERP.
The work lives in ERP, MES, CRM and document stores. An agent that only reads and writes chat messages automates very little.
Risk and legal see it last.Compliance finds out from the press release.
Under the EU AI Act, ISO/IEC 42001 or China’s data rules, discovering the requirements after the build means rebuilding.
Nobody measures it after go-live.It launched. Then everyone went quiet.
No baseline, no cost per task, no error rate. Six months later nobody can say whether it is worth what it costs.
The landing sequence
Six steps from idea to daily operation.Six steps. No theatre.
The same sequence whether the agent handles quotations, maintenance tickets or supplier queries.Boring on purpose. Boring is what survives an audit.
Pick the process
Start where decisions are frequent, the rules are known, the inputs are digital and mistakes can be reversed. Put a number on today’s cost.
Write the job description
Scope, decision rights, the tools it may use, when it must hand over to a person, and the one number it exists to move.
Connect the systems
Give it least-privilege access to the systems where the work happens, with every action logged and attributable.
Prove it on your data
Build an evaluation set from real cases, run it inside your security perimeter, and agree the pass mark before anyone sees a result.
Govern it from day one
Risk class, audit trail, human oversight and data residency, mapped to the EU AI Act, ISO/IEC 42001, the NIST AI RMF and China’s CSL, DSL and PIPL.
Run it like a new hire
A probation period, monitoring, cost per task and a scale-or-stop date. Scale what passes; retire what doesn’t.
Three ways to work together
Consulting, services, research.Advice, hands, and homework.
Consulting
Where agents pay, and in what order.Which agents are worth it. Which aren’t.
Strategy and decisions for leadership teams that want agentic AI to show up in the P&L.
- Agentic AI strategy and a ranked portfolio of processes
- Operating model, decision rights and the AI Centre of Excellence
- Governance and model risk: EU AI Act, ISO/IEC 42001, China’s CSL, DSL and PIPL
- Platform, model and vendor choices, for China and global
- Board and executive briefings
Services
From design to daily operation.Hands in the mud, not just advice.
Hands-on delivery that takes an agent from the first process map to production.
- Process selection, baselines and agent job descriptions
- Integration with ERP, MES, CRM and document systems
- Evaluation sets, guardrails and human-in-the-loop design
- Deployment in China and globally, with data residency built in
- Monitoring, cost per task and handover to your teams
Research
Independent research on agents in industry.Homework the vendors won’t do for you.
Evidence on what works, for organisations that would rather not learn it the expensive way.
- Agent reliability and evaluation in industrial settings
- China–global deployment patterns for agents and models
- Regulation tracking across the EU, China and the U.S.
- Private research briefings for leadership teams
The open questions I am working on are listed in the Laboratory. See the research agenda
Why listen to me
AI already industrialised, at group scale.I’ve done the hard part before.
- Digit-AI
The framework that takes AI use cases from pilot to production across a global industrial group, with 17.2% efficiency improvement across its China business.
Read the case study - 170+
Regulations and certifications on one security platform across 60+ countries, with the jurisdiction-specific deltas automated and compliance overhead down 40%.
Read the case study - 13
Further sites benchmarked on the Suzhou factory after my programme took it into the World Economic Forum’s Global Lighthouse Network. Templates that others copy are the point.
Read the case study