An energy company has a hundred ideas. An AI engineer has a hundred tools. The interesting question is how either gets down to one useful business. In 2024, AES and AI Fund offered an answer: AES sifted more than 100 internally sourced ideas into 12 proposals for the studio. Within six months, the partners had a funded startup, still in stealth. The narrowing was the work.
- AI Fund co-founds companies, contributing builders, recruiting and capital.
- Corporate partners bring industry problems and access to potential customers.
- The method to borrow: test a specific need before building a whole business.
That is a good place to begin with Andrew Ng’s company. Its name suggests a room where people evaluate pitches. Its daily work reaches further upstream, into the awkward interval when a problem might be a product, and a product might become a company. Someone still has to establish which.
A customer with a problem beats a room with a whiteboard
AI Fund was founded in 2017; its first, approximately $175 million fund was publicly announced in January 2018. Ng had already led Google Brain and served as Baidu’s chief scientist. Here, his technical background became part of a company-building operation: identify applications, test them, and assemble businesses around the promising ones.
The corporate relationships help explain its position in the market. A renewable-energy operator knows things about wind farms that a software team cannot learn by staring at a model benchmark. AES brings that operating context. AI Fund brings technical judgment and the machinery for creating a separate startup. Their announced priorities included battery management and worksite safety.
Mitsubishi Corporation brought another set of problems in January 2025. The initial targets were pleasingly specific: finding underused properties for small businesses in rural areas, and simplifying office-equipment leasing. You could spend a long time debating the future of intelligence. You could also make a small business’s lease less painful.
The money comes with people
In May 2025, AI Fund announced an oversubscribed second fund of approximately $190 million. Backers included Sequoia Capital and NEA alongside corporations such as HP, QBE and TELUS Global Ventures. Those are commitments to an investment fund. They describe the capital available to pursue a strategy, rather than revenue earned from customers.
The strategy is co-creation. AI Fund describes itself as a minor co-founder. Its team helps with product decisions, engineering, recruiting and company formation, then invests and supports subsequent growth. Economically, it participates through ownership in the businesses it helps create. The founder gets a working partner and shares the upside.
That places the studio somewhere a founder should examine carefully. An accelerator can help an existing team develop its company. A conventional investor can finance that team. AI Fund’s role starts earlier and extends into building. It is a closer working relationship, with more shared decisions than a cheque alone implies.
“We’re literally all builders.”Andrew Ng, Crunchbase News, May 2025
Twelve weeks to earn the next step
The current Engineer in Residence program makes the bargain concrete. A builder takes a studio-sourced idea through a full-time, on-site, 12-week sprint in Mountain View. Compensation is $10,000 monthly. Successful company formation can bring $500,000 to $1 million in pre-seed investment. Funding and the CEO role are conditional. The studio retains the work’s intellectual property if no company forms.
For someone ready to build around a selected problem, that is a defined route into entrepreneurship. Someone committed to an existing concept should read the fit carefully: the program starts from the studio’s pipeline. The useful question for an applicant is whether this particular problem and working arrangement deserve the next three months.
Startups for jobs that already exist
The portfolio makes the application focus visible. Profitmind helps retailers identify opportunities in revenue, profit and cashflow. Workhelix assesses where generative AI could help an organization and creates a tailored roadmap. Their buyers face recognizable operating decisions: what to change in a store, or where to direct an enterprise’s AI effort.
There is now an exit in the skills category, too. Pearson announced its acquisition of portfolio company Workera on September 29, 2026. Workera measures workforce skills, connecting the studio’s AI thesis to a familiar enterprise need: understanding what employees can actually do.
RapidFire AI addresses a different customer: teams customizing models. Its open-source engine, announced in September 2025, supports multiple training experiments and lets practitioners stop, resume or clone runs. The problem is the cost and delay of exploring configurations sequentially. This is software infrastructure, another part of the studio’s investment focus alongside applications.
The Taiwan hub, announced in October 2024 under Jill Shih, extends that company-building work into Asia. Shih’s background spans software engineering, product management and user experience. Those disciplines fit the task: a technically possible product still needs a workable interface, a buyer and a team capable of delivering it. Geographic reach matters when those people and opportunities sit in different markets.
The slow part moves
The AES collaboration also supplied a less glamorous lesson. Workflow systems overwrote intermediate states, losing information a model might need. AES introduced snapshots to preserve it. Better access to an industry does not automatically mean usable data. Sometimes the prerequisite for an AI product is a change to ordinary record keeping.
AES also reconsidered years-long internal incubation because of AI’s pace. Faster development, however, creates another constraint. Ng has argued that as prototyping gets cheaper, gathering user feedback becomes more important. A prototype finished this morning cannot teach much while its intended customer remains unavailable until next month.
There is a financial discipline underneath the tinkering. A studio has to turn experiments into businesses whose growth can reward investors. Useful software for a single department may deserve to exist without supporting an independent venture-backed company. Corporate enthusiasm is a starting signal; repeatable demand beyond the first relationship is the harder test. That distinction gives the market-validation work its purpose.
Builders can copy the sequence: choose a narrow problem, find people who experience it, ship a test, and use their response to decide the next change. Its usefulness depends on access to customers and evidence that the product helps. Where those are absent, faster engineering merely produces more untested software. AI Fund’s interesting bet is that company building improves when those conversations become part of the production process.