Risk builds where no one is looking. Our agents work there.

4.4 million small businesses use Xero. Most do not have an accountant — or a credit controller. The wedge is cash recovery: money already earned but slipping away because nobody chases it. The expansion is full AI bookkeeping: a conversational agent that audits, explains, and fixes the books as a subscription service.

The moat is the integration depth. Sikizana writes back to Xero through a real, permissioned tool layer — not prompt engineering — and every figure is traceable to an HMRC citation, a sector benchmark, or an explicit human approval.

No templates. No mail merge. A video that sounds like the sender wrote it for the recipient, because the agent did. The use cases are horizontal but high-value: sales prospecting, investor pitches, recruiting outreach, freelancer pitches, and conference follow-up.

The strategic value is a reusable personalization engine. Point the same agent room at LinkedIn, GitHub, Twitter/X, or any public profile, choose an intent chip, and produce a 1-to-1 asset at the cost of API calls.

One instrument, two grading oracles — both watch software commits. [markets] is graded by price (Season 1 closed, record public). [research] is graded by the dated scientific record (prospective cohort maturing). Markets were first because they are the hardest oracle: price cannot be spun, arrives in hours, and punishes errors in dollars.

The unit of proof is the call, not the trade: a directional thesis, committed on-chain before the outcome, scored against what the grading authority actually recorded. Publicly, the agent runs that record in the open. Pointing the same detectors and rubric at a customer's own repos is a leak-scan demo today, not a product — the public track record is the sales proof; the leak-scan is the direction.

Data teams spend weeks onboarding engineers to warehouse semantics. The metadata is there, but nobody reads it. DataBard turns the same analysis into podcasts, reports, and alerts so the right information reaches the right person in the right format.

Alex is the enthusiastic advocate; Morgan is the skeptical auditor. Their conversation surfaces cascading risks, ownership gaps, and stale tables in plain language. The moat is a single analysis engine with many output surfaces — every podcast, report, and alert reinforces the same underlying metadata graph, so value compounds with each new integration.

Program officers and grants managers at foundations, public programs, and corporate grant arms burn their weeks reconciling deliverables, chasing evidence, and unblocking tranches stuck in review queues. Weft sits beside the grant system of record — Fluxx, Foundant, AmpliFund, Submittable, Salesforce Nonprofit — runs a fixed checklist when a milestone is claimed, and writes a verification receipt back onto the grant record so disbursement isn't gated by manual review.

The moat is the verification loop: a fixed, auditable checklist that produces a receipt the grants team can act on without re-doing the work. Private settlement on Canton is optional pilot infrastructure — a lab for moving capital the same way verification moves — not the sales lead.

No current player combines FX-risk quantification with autonomous execution. DiversiFi makes the moat structural: a philosophy-driven values system creates identity-based retention, while the Guardian turns working-capital risk into a managed, auditable operating layer.

The expansion is from the retail saver who protects a wallet to the Ghanaian importer, the US retailer sourcing from Europe, and the UK business paying US suppliers — any company whose margins move when exchange rates do.