Senior AI expertise.Independent judgement.AI due diligence and advisory.
Strategy firms rarely have anyone who's shipped a model. Dev shops rarely see the deal. I've done both: a Cambridge NLP PhD, and an AI product taken from concept to public beta.
Cambridge NLP PhD, Oxford maths and computer science, former Head of Applied AI Research, €50m+ in AI value identified at IBM, fractional CTO taking an AI product from concept to public beta.
Who I work with
People about to commit serious money to AI, who need to know whether the plan is real before they do.
Investors and acquirers
Pricing a deal, with a term sheet, an LOI or an investment committee asking questions the data room can't answer.
Boards
About to approve a significant AI budget, and wanting an independent technical assessment first.
Leadership teams
Accountable for an AI plan or a vendor choice that has to stand up to scrutiny.
It's not a fit if you want sign-off on a decision that's already made.
How I work
An AI due diligence assessment takes two weeks, for a fixed fee, and typically starts within a week of you saying go.
Day 1
The question
We agree the question the assessment has to answer and what would change your decision.
Days 2-6
The evidence
I read the documents and the code, interview the founders and engineers, and check what the data room says against all three.
Days 7-8
The analysis
I test the findings against your thesis: what's real, what isn't, and what it means for the decision.
Days 9-10
The report
A written report with a severity rating on every finding, and a live debrief with your team.
I've taken an AI product from concept to public beta myself, so I know where production systems break and which claims to test first.
Why trust the assessment
Research-level depth, and no stake in the answer.
Your assessor
Dr Colin Kelly
I hold a PhD in Natural Language Processing from Cambridge and read Mathematics and Computer Science at Oxford. My technical work in AI spans IBM, PA Consulting and two VC-backed AI scale-ups, and predates LLMs by a decade. A former Head of Applied AI Research, I identified over €50m in AI value at IBM across telecoms, financial services and automotive. My recent work includes an organisation-wide AI strategy for a major global philanthropic foundation, and 18 months as fractional CTO taking an AI product from concept to public beta.
Read my storyNo stake in the answer.
An assessor who profits from finding work has a reason to find it. I take no build or remediation work on anything I assess, charge a fixed fee, and have no vendor affiliation.
Direct access, not delegation.
When you engage Agathon, you work with me. I don't sell then disappear. I don't hand you off to juniors. The person on the strategy call is the same person doing the work.
Recent work
Strategy, advisory and fractional CTO engagements, with the same technical judgement I bring to AI due diligence.
Case study
Fractional CTO Building Patent-Pending AI Product from Zero to Beta
How we took a ghostwriting firm from copy-pasting into ChatGPT to a patent-pending AI product in beta with enterprise users.
20 months
Engagement duration
5 hires
Team recruited
85%
AI rendering speedup
Case study
AI Strategy and Implementation Roadmap for a Major Global Foundation
How we helped a large communications division turn AI momentum into an implementation plan — in six weeks.
6 weeks
Engagement duration
11
Stakeholder interviews
2 delivered
Technical blueprints
Case study
AI Leadership Advisory: Transforming a Communications Firm's AI Capability
How we turned fragmented AI experimentation into shared frameworks, tested playbooks, and a concrete product roadmap over two months.
2 months
Engagement duration
3 half-day
Workshops delivered
Full team
Team members trained
How I can help
Independent AI due diligence when there's a decision on the table, advisory that grows your own capability, and part-time technical leadership for a small number of companies at a time. I lead each engagement myself, from first conversation to final deliverable.
Lead service
AI Due Diligence
Independent technical assessment for investors and boards of whether the AI is worth the price, and whether the advantage will last. Before you commit capital.
Learn moreAI Leadership Advisory
Strategic advisory for leaders building genuine internal AI capability -- the teams, processes, and judgment to act without outside dependencies.
Learn moreFractional AI CTO
Part-time AI CTO or Head of AI with genuine decision authority, for a small number of companies at a time.
Learn moreIncluding strategic workshops and training
Strategic workshops
Intensive, practical sessions designed to build real capability, not just awareness.
Latest insights
Questions to ask an AI vendor before you sign
Before signing an AI vendor contract, replace trust in the rehearsed demo with written questions about the system's workings, data, dependencies and exit terms, scored for specificity.
Big 4 consultant vs independent AI advisor: who to trust with AI strategy and due diligence
Big 4 firms sell scale, insurance and reassurance while independent advisors sell unconflicted senior judgement, and since AI due diligence is essentially the purchase of scepticism, the deciding factor is not the advisor's intelligence but their incentives.
Why I make clients pay for the part everyone wants for free
Cheap generation made working prototypes trivial; knowing what deserves to be built remains the only job that matters.
What "sovereign AI" actually means, and when it matters
A clarifying primer that recasts "sovereign AI" not as a slogan but as control across four layers - data, compute, weights, governance - and shows that only a small, identifiable minority of enterprise workloads genuinely need more than the data layer existing tools already provide.
Inside sovereign AI governance: How it works when the frameworks meet reality
Sovereign AI governance fails not at the framework level but in the gap between documented controls and operational reality, where jurisdictional contradictions, supply chain dependencies, and untested incident response processes accumulate risk that compliance paperwork cannot see.
We've deployed Microsoft Copilot 365. Now what?
Deploying Microsoft Copilot is a logistics problem; extracting value from it is an organisational one, and most companies confuse finishing the first for starting the second.
Securing AI agents: Why tool use creates new attack surfaces
Securing AI agents requires treating the surrounding architecture as the threat surface, not the model itself, because authentication gaps, over-provisioned tool access, and prompt injection vulnerabilities combine to make your most capable agents your most dangerous ones.
Let's talk about the decision in front of you
Whether there's a deal to assess, a vendor to evaluate or an AI plan you're accountable for, the first step is a conversation. I'll tell you quickly if it's not a fit.
Dr Colin Kelly