I work with teams building with AI and agentic systems -- from early-stage startups to established organizations -- who have a real problem and a sense that these tools should be part of the answer, but who don't yet have the in-house depth to design the system, judge the trade-offs, or build the first version.
I'm most useful when the stakes are technical and the path is still uncertain -- when you need someone who has built these systems to tell you honestly what's feasible and what the smallest valuable first step looks like.
Alongside the consulting, I take a smaller number of research and advisory engagements on verified geospatial systems -- provenance, integrity, and trust in spatial information. That practice sits next to a lot of current AI safety work.
Where AI genuinely earns its place in your product or operations, and where it doesn't. A clear, honest view of what to build, in what order, and why.
Where in this would a language model genuinely change the answer, and where would a classical system be cheaper and more reliable?
I help you and your team build a clear picture of what's actually happening in AI and why it matters for your work, so you can decide from understanding rather than from pressure.
Which capability claims should we take seriously this quarter, and which can we wait out without falling behind?
Architecture for systems that use language models to reason, plan, and act -- tool use, retrieval, evaluation, and the guardrails that keep them reliable.
How do we keep an agent's tool use reliable when the long tail of production looks nothing like our eval set?
Working prototypes, quickly -- so you can put something real in front of users or investors instead of arguing about a slide.
What is the smallest version of this that would actually teach us what we need to know next?
An assessment of the technical substance of a product, team, or codebase -- for founders sharpening their own thinking, or investors who need a second opinion.
Is this product genuinely model-bound, or is the demo doing most of the work?
A part-time technical lead or AI architect, for when you need senior judgment in the room but not a full-time hire yet.
Who in our room can push back when a vendor or a model lab tells us what their roadmap will deliver?
one to two weeks · fixed fee
A written diagnosis and a roadmap -- what the real problem is, what's feasible, what to build first, and what to ignore for now. Concrete enough to act on or hand to a developer. The natural first step when you know AI should help but aren't yet sure how.
one to two days a week · three-month minimum
Ongoing senior technical judgment -- architecture decisions, code and design review, prototyping, hiring input, and acting as the AI lead in the room. The deliverable is momentum and good decisions, not a single artifact. For when you have real work in flight and need someone senior shaping it every week, but a full-time hire is premature.
scoped outcome · agreed price
The thing itself -- a working prototype, a production integration, a due-diligence report -- built to an agreed specification. For when the problem is well enough understood, often after a discovery engagement, that we can name the outcome and fix the price.
These aren't rigid, and they aren't priced off a menu. We almost always start with a short call to work out which one -- if any -- actually fits, and what a fair scope and price look like for your problem.
If you would like a sense of how I think about this work before we talk, I put together a short deck -- Building with agents -- a pass through foundational concepts in agentic systems, to help build an accurate mental model.
Drop your email and I will send you the link.
I'm a geospatial technologist and researcher based in London. I co-founded Toucan Protocol, spent time at Ordnance Survey, hold an MSc from UCL's Centre for Advanced Spatial Analysis, and am a Research Affiliate at the University of Maryland's Department of Geography.
Geospatial and location intelligence are my specialism, not a limit. The work itself -- finding where AI fits, making it reliable, building the first version -- applies well beyond them.
If any of this sounds like your problem, I'd like to hear about it. The right starting point is almost always a short conversation -- tell me what you're working on and where you're stuck.
john@johnx.co