Come with capability, not certainty.

Overture Residency is seven intensive days. This application evaluates the person before the startup. You do not need a company, a cofounder, a deck or a finished idea. You do need the capacity to apply AI to a real problem, and the willingness to change your mind when the evidence changes.

We read for how you operate. What you have built, sold or run when it was difficult. Whether you can name a constraint in the world, not a category. Whether you will show up for seven days and try to form a company, not perform one.

If there is a fit, we write back. If there is not, we will not string the process out. Adjacent problems can enter. Vague impact cannot. Read the four project types below, then apply.

Four streams / South Africa

The project types.

Overture forms companies across four streams. Each is AI applied to a concrete problem — not the category on its own. You do not need a finished idea. You need the capacity to go after one of these, or an adjacent problem that is equally specific. Read them before you apply. The application asks which of these you are prepared to work on.

01 / AI for Work

Unemployment in South Africa

South Africa has one of the highest unemployment rates in the world. Capable people sit outside the labour market. The usual response is a programme, a placement scheme or a training pipeline. Those rarely create work people will actually take.

The wedge

We are not a jobs company. We apply AI to make work exist, get done and get paid — matching, tasking, supervision, quality and demand — at a cost the market can bear.

Company shapes

Demand: AI that lets a business or household buy labour they currently cannot manage — scheduling, trust, quality, payment.
Supply: AI that turns informal or fragmented work into something a person can do and get paid for.
Creation: products that generate work the market did not have, rather than shuffling CVs through another board.

Not this

Job boards. Learnership mills. Youth-employment branding. A marketplace with no buyers.

Evidence

Someone is hired or paid who was not. A buyer gets work done they were not getting done. Repeat, not a pilot.

02 / AI for Transition

Re-education and job change

People need to move from one kind of work to another: from a dying industry, from campus, from informal work into something that pays. The current path is slow, expensive and often dishonest about what happens after the certificate.

The wedge

We are not a university, a bootcamp or a content library. AI should collapse the time, cost and fiction between the person they are and the work they can do next.

Company shapes

Assessment that is honest about what someone can actually do next — not a personality quiz.
Training tied to a specific job, where the company only wins if the person moves.
Tools that let a worker keep earning while they change, instead of dropping out of income to study.

Not this

Another LMS. Generic upskilling. Certificates nobody hires for. A course catalogue.

Evidence

A person is in different, better work. Time and cost beat the incumbent path. Someone paid for the outcome.

03 / AI for Connectivity

Access to the internet

Bandwidth, devices and cost still decide who can participate. Products designed for always-on fibre and flagship phones fail where most people are: prepaid, shared devices, intermittent signal, expensive data.

The wedge

We are not an ISP or a hardware reseller. AI is used to make scarce connectivity more useful — compression, offline-first, shared access, last-mile operations — or to build products that work on the network people actually have.

Company shapes

Software that completes a real task on prepaid, shared devices and intermittent signal.
AI that stretches existing capacity: caching, compression, local or delayed inference, sync when the network returns.
Last-mile operations — community access, device sharing, agent networks — made viable by AI ops, not another coverage slide.

Not this

A fibre or 5G pitch. An app that only works on good Wi-Fi. “We will partner with a telco” with no product.

Evidence

People who were poorly connected complete a real task. Cost per useful session drops. It works on the device and tariff they already have.

04 / AI for Local Market Productivity

How small businesses scale

Most local firms never become companies that can grow. The founder is the accountant, the salesperson, the compliance officer and the operations team. Formalising and scaling still costs too much relative to the business. Credit, payments and working capital are already crowded. That is not this stream.

The wedge

We apply AI to the operating system of the local firm: formation, customers, books, compliance, hiring, inventory. The analogue is Stripe Atlas — not for Delaware startups, but for a spaza, contractor, salon or trader who needs to become a real, scalable business.

Company shapes

Formation: register, tax, contracts, identity — the local Atlas. A real company in days, not months of offices.
Operations: invoicing, stock, customers, staff and scheduling run by AI so the founder is not the entire company.
Demand and records: getting customers without a marketing team, and keeping books that unlock a supplier, a tender or a lender — without becoming one.

Not this

Another SME lender. A neobank. A payments app. A financial-inclusion wrapper. A dashboard nobody operates.

Evidence

A local business that could not scale now does more volume, with less founder time, as a more real company. A registration, a customer, a transaction — not a demo.

Adjacent problems can enter if they are equally concrete and AI is the method. Vague “impact” is not a stream. If none of the four is yours, say so in the application — and name the constraint in the world, not a category.

The application

Tell us who you are.

Short answers. Concrete ones. Links help. The idea field is optional on purpose.