Twelve physical-world businesses two AI-native engineers could start — where software expertise is the unfair advantage, and the moat is something a vibe coder can't clone.
A single large client, invoiced monthly. Comfortable, precarious, and going nowhere. CTOs-for-hire in a shrinking market.
Frontier models and coding agents let laymen ship software. Demand for hired expertise dries up month by month. Two years out is anyone's guess.
Anyone semi-technical with an agent can clone a SaaS in weeks. Shipping software is no longer a defensible skill — it's table stakes.
Enter proven, boring markets where incumbents are non-technical and slow — then run the entire operation on agents: lead capture, quoting, scheduling, dispatch, comms, compliance, back office. The moat is atoms — assets, leases, licences, contracts, reviews, brand, capital. The margin is agents. Each layer of automation makes the next site or customer cheaper for us, but not for a copycat.
Demand already exists. Incumbents are non-technical and slow to adopt AI. We win by executing better, not by inventing demand.
The cost and quality problems live in operations, not product — exactly where agents create an unfair advantage.
Defensibility comes from assets, licences, location, contracts and trust — things that can't be forked from a repo.
Automation makes every additional site, job or customer cheaper for us — and only us. Margin widens with scale.
Proven markets, cashflow in weeks, fundable from savings. Boring is a feature — smart founders avoid these, which is exactly why we shouldn't.
Capital: within £20k · Ideas 01–04Pick one high-ticket trade — drainage, locksmithing, gutter & roof maintenance, EICR/gas compliance, end-of-tenancy cleaning. Employ or subcontract the labour; own everything else. Local jobs are won by whoever answers first and quotes fastest — incumbents miss calls and quote in days. Our agent answers every call and WhatsApp in seconds, 24/7, prices from photos, schedules, dispatches, chases reviews, invoices.
Response time is the #1 conversion lever in local services — and AI just made instant response free.
Google reviews compound, local brand, vans, trained staff. None of it clonable from a laptop.
£20k covers vehicle, kit and launch marketing for most trades. Cashflow-positive in weeks.
You're managing tradespeople — hiring and quality control is the real job.
Same thesis, different entry: acquire a small UK services business (£50–150k, often part seller-financed, retiring owner) with existing customers and cashflow — then AI-transform its operations. The £20k becomes a deposit. Skips zero-to-one customer acquisition, the hardest part, and applies our edge to a machine that already works.
Retiring boomer owners with no succession plan — an unusually deep buyer's market for micro-acquisitions.
Existing contracts, customer base and reputation from day one — plus everything we automate on top.
Do it twice, prove the ops stack transfers, and it becomes a roll-up story for investors.
Diligence. Small businesses hide problems; the owner often IS the business.
Lettings and block management: recurring monthly revenue per unit, universally hated incumbents, and 80% of the job is admin — contractor coordination, compliance deadlines, service charges, tenant comms. All of it agent-shaped. Variation: short-let / Airbnb management — faster to win clients, more ops-intensive (which favours us), 15–20% of rent.
Growing compliance burden is drowning small agencies in exactly the admin agents eat for breakfast.
Management contracts are sticky, trust takes years to copy, and there's real physical presence — inspections, keys, contractors.
Near zero. Grows linearly and predictably; every unit is recurring revenue.
Slow client acquisition; landlords switch agents reluctantly.
Smart fridges and micro-markets in gyms, offices and residential buildings. AI runs demand forecasting, dynamic assortment per location, restock route optimisation, and telemetry-driven maintenance. Each good site signed is one a competitor can't have.
IoT hardware is commodity-cheap; the winners are decided by forecasting and routing software.
Exclusive site contracts and physical footprint. Location deals compound.
£20k buys a starter fleet of smart fridges.
Thin margins until scale; a grind of site-by-site sales. Better as a second engine than the main event.
Real software to write, real assets to own. These take longer to stand up — and are proportionally harder to copy.
Capital: £20k–£80k · Ideas 05–08A local CNC / laser-cutting / 3D-printing job shop where customers upload a CAD file and get a price in minutes — against an industry norm of days and phone tag. Xometry's insight, but owned and local, serving small manufacturers, product designers and trades. The quoting engine is real software; the machines are a capital moat.
Incumbent job shops are run by machinists, not software people — the quoting/scheduling layer is where the margin leaks.
Machines, capacity, certifications, sticky repeat B2B customers.
£20k of entry-level machines validates demand; investment scales it.
Same model for signage / large-format printing — less sexy, faster market.
24/7 access-controlled spaces run entirely by software: golf simulator studios, podcast & content studios, rehearsal rooms, dog exercise fields, self-service gyms. Booking, entry, payment, computer-vision monitoring, cleaning coordination — all automated, near-zero staff cost. Inherently multi-site: prove one, raise for five.
The unit economics only work with deep automation — which is precisely our skillset, and the moat itself.
Lease, fit-out, and an ops stack a non-technical operator can't replicate.
£30–80k per site depending on fit-out.
Best blend of realistic and fun for two sports-and-tech founders. Golf sims especially.
Run recreational leagues — 5-a-side, netball, padel, pickleball — renting existing facilities. Agents handle the genuinely painful parts: scheduling, skill-based matchmaking, team-filling, comms, standings. Community is a real moat; people stay for the people. Second act: once we own demand (hundreds of paying players), open our own venue with pre-loaded utilisation.
Racquet-sports and social-league demand is booming; scheduling admin is why most operators stay small.
Community and network effects locally; then a physical venue funded by proven demand.
Near zero to start — court hire is paid by player fees. The venue comes later, de-risked.
Strong personal fit — and a runnable weekend experiment alongside anything else.
Roof surveys, gutter inspections, solar panel checks, site progress monitoring — sold B2B to property managers, insurers, surveyors and solar installers. The AI angle is real: agents generate the inspection report from imagery automatically. Report-writing is where incumbents burn their hours.
Drone hardware is cheap and capable; automated reporting turns a day of desk work into minutes.
CAA GVC certification, B2B relationships, and a report pipeline competitors would have to build.
Kit and training comfortably under £20k. High margin per job.
Smaller market per city — a great business, maybe not a huge one.
Bigger capital, bigger prize. The capital requirement is itself the moat — nobody enters these spaces with a laptop and a weekend.
Capital: raise required · Ideas 09–12Padel is still supply-constrained in the UK. AI-native operations: dynamic pricing, skill-matched open play (the killer feature — solving "who do I play with" drives utilisation), automated leagues, coaching content, near-zero front-desk staff. Incumbent operators are property people, not product people — utilisation and community are software problems.
Fastest-growing racquet sport in Europe; UK court supply still lags demand badly.
£300k–£1m+ site capital, planning permission, and community lock-in.
Combine with idea 07 — own the player community first, then the venue, de-risked for investors.
Capital-heavy; site acquisition and planning are the hard, slow parts.
Agents continuously scan planning portals, Land Registry, auction catalogues and sold-price data to find mispriced small sites and permitted-development opportunities — faster than local developers working off relationships and gut feel. Start as deal sourcers/packagers (fee per deal, near-zero capital); graduate to taking positions with investors.
The data is public but messy and distributed — a genuinely durable software edge nobody local has.
Property is the ultimate physical moat; the sourcing engine compounds with every deal.
Near zero as a packager; investor capital when taking positions.
Slow feedback loops; planning risk; deals measured in months, not weeks.
The meta-play: take idea 01, 06 or 07, build the "agentic operating system" for that business as deliberately reusable infrastructure, prove it at one or two sites — then scale via owned sites (raise) or franchising/licensing the ops stack + brand to operators. Most franchise systems are binders of PDFs; ours would be software that runs the business.
Agentic ops stacks are brand new — no franchise system on earth ships one yet.
Brand + a living software system franchisees depend on daily. Recurring royalty revenue.
Hold as the explicit endgame regardless of which entry we pick — build everything reusable from day one.
Only works after a proven, profitable unit. It's act three, not act one.
Massive structural demand, chronic labour shortage, and the binding constraint is rostering, compliance and coordination admin — heavily agent-shaped. CQC registration is a real regulatory moat. Operationally and emotionally heavy, reputationally unforgiving — only if we want a mission-heavy decade.
Ageing population, structural undersupply, and incumbents running on spreadsheets and phone calls.
CQC registration, council contracts, carer workforce, family trust.
Arguably the largest market on this list.
People's lives depend on the roster working. Zero tolerance for failure.
| Idea | Capital | Time to cashflow | Moat depth | AI edge | Personal fit |
|---|---|---|---|---|---|
| 01 Home services | ≤ £20k | Weeks | |||
| 02 Buy a business | £20k + finance | Day one | |||
| 03 Property mgmt | ≈ £0 | Months | |||
| 04 Smart vending | ≤ £20k | Months | |||
| 05 Microfactory | £20k → raise | Months | |||
| 06 Unstaffed venues | £30–80k | Months | |||
| 07 Leagues → venue | ≈ £0 → raise | Weeks | |||
| 08 Drone inspection | ≤ £20k | Weeks | |||
| 09 Padel club | £300k–1m+ | Year+ | |||
| 10 Property dev | ≈ £0 → raise | Year+ | |||
| 11 Franchise endgame | After act one | Year+ | |||
| 12 Home care | £20k+ | Months |
Launch 01 home services or 03 property management with the £20k. Revenue in weeks, and every agent we build becomes reusable infrastructure.
Run 07 social leagues at near-zero cost. It's fun, it's our world, and it's the cheapest possible option on a future venue.
With revenue, a proven agentic ops platform, and a player community, raise for 06 unstaffed venues or 09 padel — with 11 as the endgame.
The trap to avoid: six months picking perfectly instead of eight weeks testing cheaply. The ops stack transfers — an early "wrong" pick costs less than it looks.
Validate locally: mystery-shop competitor response times · scan acquisition marketplaces · map court & venue supply · price the kit. Then commit to whichever one pulls hardest.