From functional capital tointelligent assets
In the generative-AI era the moat is no longer how thick your IT stack is. It is the density of structured functional capital. We do not sell heavy, hollow general-purpose systems; we turn the industry knowledge your firm has accumulated for years into visibility and execution that hold up in front of a large model.
Two things happening right now
More systems, the same manual work
Companies have bought plenty of systems, yet the steps that cost the most people — approvals, checks, roll-ups, replies — are still carried between systems by hand. Another heavy SaaS is just one more system someone has to feed. We find those steps first and rank them by cost and by whether AI can take them over.
See the cases →The cost of moving tacit knowledge
Compliance review, due diligence and content distribution run on individual experience; off-the-shelf SaaS cannot fit non-standard work, so output per person hits a ceiling. Experience that stays in one person’s head cannot be called, copied or audited — the most expensive cost on the books, and the least measured.
Not a system purchase — end-to-end delivery
Our agile build is run by a cross-disciplinary group of expert models, on a lightweight, low-code base, delivered end to end and judged on results.
- 01
Your workflow stays as it is — the system grows onto your process, not the other way round.
- 02
Plug in and run — prove one high-drain step first, then extend sideways. No six-month IT programme.
- 03
Paid on outcomes — every stage ends in something you can accept or reject, not a proposal nobody opens.
Theseus · the FDE system behind NeuronSpark
Theseus is the AI forward-deployed engineering system NeuronSpark developed for its own work. The name comes from the Ship of Theseus: the business keeps sailing while we replace one plank at a time (a form, a manual approval, a heavy piece of software) with an agent or an automated workflow. It stays the same ship, and every plank is better than the one it replaced.
- 01
Purpose first
Each plank states why it exists, which business goal it serves, and how that will be measured.
- 02
Measure before and after
Every swap is measured before and after. If no difference can be measured, the swap is not finished.
- 03
Irreversible steps stay with people
Every tool is marked as reversible or not. An action that cannot be undone needs a person's approval.
- 04
Every run is recorded
Success, failure and "did not run" are recorded separately, so every swap can be traced.
Reverse-engineering functional capital
Your know-how is sealed into a system that is yours, in three layers:
Cognitive slices and a knowledge graph
We leave your workflow alone and take what you already have — archives, past case files, content library — then build a knowledge graph that is specific to your industry through cross-disciplinary semantic mapping.
A light interface (MVP), built at pace
No six-month development cycle. At hackathon pace we ship a minimum viable product with a dashboard people can read. Zero learning curve: your staff call complex analysis in plain language.
A data flywheel that keeps calibrating
The system ships with a feedback loop. Every correction and judgement your staff make tunes it closer to how your business actually works.
A five-layer delivery stack
One skeleton, different functional-capital slots: insert a financial-audit skill capsule for a family office, swap it for a cross-border news-editing capsule for a media group — the frame itself never changes.
- L1
Agile interaction plane
Agile Interaction PlaneNo heavy front end. Instead of a comprehensive system, an AI coding stack produces a light dashboard or workbench aimed at one specific pain. The interface is the pain radar: high signal-to-noise visuals meant to make the problem legible in the first week.
- L2
Cognitive orchestration brain
Cognitive Orchestration BrainA decision hub abstracted away from any one business: task parsing (intent, plan, constraints) and orchestration. Not prompt plumbing — non-standard requests become a standard task tree that dispatches the skills below.
- L3
Functional capital capsule
Functional Capital CapsuleThe soul of the system: a senior expert’s intuition and undocumented experience, fixed into reusable atomic skill capsules and proprietary knowledge-base calls. This layer is a modular slot — change industry by changing the slot, not the frame.
- L4
Heterogeneous connectors
Heterogeneous ConnectorsNo energy spent on legacy plumbing. Semantic connectors turn unstructured black-box data into a structured network a model can read. Business connectors reach ERP, CRM and OA. Remote controllers use RPA where a compliant intranet or legacy asset exposes no API.
- L5
Sovereign asset base
Sovereign Asset BaseThe actual moat. On top of data governance, a proprietary RAG vector graph accumulates; the longer the system runs, the smaller its predictive error for this particular firm, until the base cannot be separated from it. Fine-tuned vertical weights are an advanced option, taken only when needed.
Note: L1–L4 and the RAG graph in L5 are what we deliver today; fine-tuned vertical weights are an advanced option, not part of standard delivery.
Which work we take on
The same effort spent on different work differs by an order of magnitude in what it leaves behind. We screen on two axes — how much tacit knowledge the industry holds, and how reachable its data is — and we use the same grid to tell you when not to do this at all.
Low tacit knowledge × reachable data
Clear rules and standard processes, already solved well by mature SaaS. We will tell you to buy the existing product; building it adds nothing.
High tacit knowledge × reachable data
Decisions lean on a senior expert’s intuition and cross-disciplinary judgement, and the data can be reached. Effort converts directly into the client’s sovereign asset. Nearly all of our work sits here.
Low tacit knowledge × closed systems
Linear rules with little judgement to accumulate, and delivery would mean rebuilding legacy back ends. That is traditional systems-integrator territory, where AI adds no emergent value. We do not take it.
High tacit knowledge × closed systems
The experience is valuable but nothing exposes an API, or the records sit inside a compliance intranet. Worth doing — but the first phase’s budget goes to L4 connectors and RPA takeover, not to the agents.
This grid is also how we say no. We concentrate on the sweet spot because that is where every unit of effort settles into your asset rather than becoming a one-off engineering bill.
Short steps, frequent proof
Never a long IT swamp. Each phase ends in a concrete output, and you can stop at the end of any phase.
- Weeks 1–2
Find the target
Pin down the step of the business that costs the most human effort and suits AI best.
Output: a tailored Structured Functional Capital Assessment
- Weeks 3–4
Ship the prototype
The first vertical agent interface — a due-diligence assistant, or an automated approval flow for one step.
Output: a test that runs in your real environment, with a measurable change
- Months 2–3
Extend and harden
Extend sideways to more nodes and join the separate agents into one working flow.
Output: your own AI moat, closed as a loop
What we have built
This is a path we have walked, not a proposal: the platform we developed for MEMECMO, Musicraphy, DirectorAI, a live-biologics brand system, and a CRM with sales agents for a Vietnamese out-of-home media group.
See the full list →Start with one diagnosis
Two weeks, one Structured Functional Capital Assessment. Decide whether to continue once you can see the findings.