The bankruptcy recommendation was already on his desk when Matt Leta saw the challenge that would change his year. In March 2023, a Twitter thread proposed the HustleGPT experiment: let AI direct you, follow its instructions, and try to build a business in 30 days. Leta had just watched his previous agency post its best revenue year ever before an audit revealed collapsing collections and mounting taxes. He had $100 left. He decided that killing the company to protect his bank account would kill the version of himself he wanted to be. He says he chose instead to repay creditors from future profits while rebuilding the business.
Three years later, Future Works describes itself as an AI-native transformation firm serving enterprise clients through 12-week cycles that combine AI agents with named experts and outcome-linked fees. That company started as a $100 experiment documented daily on a public thread that eventually reached 23,900 followers.
People are also reading…
The PartnerGPT reframe
The original HustleGPT challenge asked participants to follow AI instructions and act as its hands in the world. That framing did not make sense to him. An AI cannot take accountability for a bad call, feel it when a client is unhappy, or pick up the phone and repair a relationship. Leta called his version the PartnerGPT challenge and treated GPT-4 as a partner, not a boss.
He briefed a dedicated session on the business, treated its responses as input rather than instruction, and pushed back when the logic did not hold. It functioned, as he wrote at the time, as "better than assistant, but not strong enough to be a boss." The AI even named the company. "Future Works" came back in one of those sessions. It fit. He kept it.
The 30-day scorecard
According to Leta’s account of the experiment, by the end of 30 workdays it had generated $1.32 million in total pipeline, including $556,000 that he described as closed or having a 95% probability of closing. The account reached 23,900 new social media followers starting from zero. The model behind it was simple: small packaged projects offered to fast-moving buyers at 70 to 90% below what a traditional firm would charge, then built from there. No full-time staff were hired during the challenge. Contractors came in only after clients were signed.
Two things did not work: the existing communities and mailing lists from his previous agency yielded zero projects, and a self-serve ordering system went nowhere. Everyone still wanted a call. The relationship is what the tool ran on.
Who carries the accountability in agentic AI consulting
Every call during those 30 days was Leta's. The AI helped him think more clearly, move faster, and test ideas he might have talked himself out of. The execution and the risk belonged to him. That is the operating principle that may make the model work at scale. Named experts own the decisions that matter. AI agents handle the throughput that used to consume weeks. As the team scaled from one person to a hundred, the accountability structure stayed the same.
From $100 to outcome-staked enterprise AI transformation
The firm signed its first Fortune 500 client before the end of month six, with no outside investors. The outcome-staked model that governs every engagement today started as a practical necessity. "Tying fees to results was the only honest way to ask someone to take a chance on us," Leta said. Leta argues that linking some fees to delivered outcomes can better align incentives between the firm and its clients. According to Leta, the firm later crossed $2 million in revenue and added Fortune 100 clients. Leta says the experiment began with a $100 budget and no full-time staff, with employees added only after clients were signed.
The full playbook behind the model is public at future.works.
