Proposed — the lab is being built
The R&D Lab
Most companies guess what to build next. We run agents whose only job is to find out: they read the audit gaps, listen to what customers keep asking for, watch the market — and score every idea in hours and money before a human ever sees it.
Honest by design. Everything on this page is proposed. The idea agents are specified, not running yet. No idea here is a product until it passes the scoring, the human approval, a written spec, and a real pilot. Scores are shown when an idea has actually been scored — never invented to make a list look finished.
The idea agents — three jobs, one lane each
They are part of the research and development team. They find ideas. They never build anything themselves, spend money, or contact a customer. Finding ideas and building products are separate jobs, kept separate on purpose.
Scout
Watches the market. Competitor launches, new tools and models, what other AI consultancies are selling — and the gaps nobody owns: diagnose-before-prescribing, one brain over many tools, a proof layer that measures results, and the trades and real-estate verticals.
Miner
Reads our own evidence. The lowest-scoring sections across completed audits, the questions customers ask over and over, the jobs that keep coming up on sales and support calls. A pain that repeats is a product waiting to happen. Audit answers are used in aggregate only — no single customer's answers are ever shared or used to sell to someone else.
Valuator
Prices every idea in time. Hours saved per week × the owner's Buyback Rate = time value per year. Build cost estimate next to it. Then the 4× test: the idea must return at least four times its yearly cost, or it is shelved — not dressed up.
And the Builder. A fourth agent turns an approved spec into a working agent — by assembling it from vetted workflow templates, never inventing from scratch. It is trained on the template library and the acceptance tests, and every build it makes is logged. See the Agent Builder to try the flow.
How an idea becomes a product
Idea
Scout, Miner, or Valuator proposes a one-page idea card: the job, the evidence, hours saved, time value, build shape, and what it is not.
Scored
The idea is scored 0–5 on five criteria (below). Anything under 3.00 out of 5.00 does not reach the spec stage.
Spec
A human approves the idea. Then a written spec: purpose, users, inputs and outputs, tools, acceptance test, and what it must never do.
Pilot
The Builder assembles it from the template library, it passes its acceptance test, and it runs as a 7-day trial with a real customer.
Product — or destroyed
Pilot results decide. A product converts to paid production. A failure is destroyed — and recorded with the reason, so it is not proposed again without new evidence.
The loop learns
Every build — good or bad — is logged. A build that works for many customers becomes a new template, after a human review. A failed build becomes a lesson attached to its template.
How ideas are scored
Five criteria, 0–5 each, weighted. The score is a recommendation, not a decision — a human still approves every idea before it moves.
| Criterion | Weight | What earns a 5 |
|---|---|---|
| Time value | 30% | Saves 10+ hours a week for a typical customer at a $50+/hour effective rate. |
| Demand evidence | 25% | The same pain shows up in audit gaps and in customer conversations — not one or the other. |
| Build fit | 20% | A vetted template covers 80% or more of the job; no new infrastructure needed. |
| Strategic fit | 15% | Strengthens the Master Brain — adds a worker, a lane, or a vertical — rather than a one-off. |
| Risk | 10% | Low data sensitivity, no new compliance surface, the approval gate already covers it. |
Prioritisation rule. Highest weighted score first; ties broken by time value. Ideas under 3.00 do not reach the spec stage.
Current ideas — the starter list
All proposed — none is a product yet
The ideas the lab starts with. Scores are shown only where an idea has actually been through the scoring — the rest are marked as being scored, because inventing a score would be the opposite of what this lab is for.
CallGuard
AI receptionist — answers every call, books the job. The flagship.
Where it came from: audit gaps (missed calls, slow replies) and customer conversations. Status: build spec complete, 12 acceptance tests written — awaiting its first pilot.
LeadRelay
Fast follow-up — replies to new leads in minutes, not days.
Where it came from: speed-to-lead gaps in audits; the most repeated customer pain after missed calls. Status: build spec complete, 12 acceptance tests written — awaiting its first pilot.
Master Brain / Master Chief
One brain over many tools: receptionist, follow-up, quoting, and booking under one orchestrator, one business memory, one governance layer — with CallGuard as its first worker.
Where it came from: the market gap nobody owns, and the event-driven architecture work. Status: architecture designed — being scored.
Real Estate Master Brain
The master brain for a whole brokerage: lead capture, showings, follow-up, and reviews, with team-lead and owner dashboards. Proposed pricing: $299/month for five agents, plus $29 per extra agent.
Where it came from: a vertical nobody owns yet. Status: page live, pilot-first — being scored.
Proof Dashboard
Measures every agent in plain numbers: calls answered, leads recovered, hours returned, cost per outcome. The evidence layer that turns “try AI” into “prove AI”.
Where it came from: the measurement gap — customers cannot see what their agents actually did. Status: idea only — being scored.
Contractor Brain
One brain for a contractor: job costing and cash-flow forecasting, workforce tracking, and an AI email manager that prioritises the inbox — starting from the accounting software they already run.
Where it came from: audit gaps in trades businesses and the contractor-software research. Status: idea only — being scored.
The idea agents themselves
Scout, Miner, and Valuator — the agents on this page. The lab's first product is the lab itself: agents that invent the next agents.
Where it came from: the need to stop guessing what to build next. Status: specified — this page and the builder spec are the spec.
The other ten systems
ServiceDesk, DealCoach, SchedulePilot, QuoteReady, Retain360, KnowVault, CashFlow Watch, CrewCoach, Reputation Loop, and AIMBEX Studios — all specified with acceptance tests, all awaiting their first pilot.
Where they came from: the competitive rebuild work. Status: specs complete — queued behind the first pilots.
Have an idea for us?
If there is a job your business keeps losing hours to, tell us — that is exactly what the lab is for. The best ideas in the list above started the same way: a real job, described in plain English.