The first version of Amigo existed as a bill before it existed as software. Ali Khokhar had left Upwork a few months after ChatGPT arrived, convinced that AI would reshape marketplaces built around human labor. He had no co-founder and no working product. What he did have was a hypothesis: people who sold expertise might pay for a digital version of themselves, one that could keep answering questions after the expert logged off.
Compliments were easy to collect. Khokhar wanted a harder signal, so he asked prospective customers for a $499 deposit. The product was not ready. The promise was specific enough that people paid anyway. By his later account, those deposits and an early pilot produced about $12,000 before a line of production code had been written. The number was modest. The transaction changed the meaning of everything around it. Interest had become commitment.
His outreach had its own small hack. Messages to coaches were going unanswered until he added a postscript: “I am also looking for a coach myself.” The reply rate, he said, moved from roughly 2 percent to 60 percent. The line worked because it reversed the posture. He was no longer only asking for attention. He might also become a customer.
Learning to make the question testable
Khokhar's route to that moment did not begin in a lab. He graduated from Queen's Commerce in 2018. An HSBC International Business Award helped fund an exchange at Università Bocconi in Italy, where he worked with classmates from around the world and explored an early interest in working in Europe. He then joined Google, first in product marketing on the core ads business and later on a consumer growth team. The jobs put him close to the machinery of distribution: how a product finds people, how behavior changes and how growth becomes a system.
At Upwork Labs, the company's internal incubator, he moved into product leadership and worked on AI-native ideas tied to the future of work. The perch was unusually relevant. A labor marketplace could see both sides of the coming change: AI as a tool for workers and AI as a potential substitute for some of the work being bought. Khokhar left in 2023 to test that tension directly.
When a coaching company wanted to see the early Amigo concept, there was still little to show. Khokhar cloned the prospect's voice, manually scripted a conversation and assembled a video that behaved like the product he wanted to build. The demo won a $10,000 pilot. It was a piece of startup theater with an honest purpose: compress months of building into a concrete buying decision. The prototype did not prove the system could scale. It proved the customer cared about the experience.
“I am also looking for a coach myself.”The postscript that changed an early sales conversation
A couch, 100 pitches and a co-founder
The first fundraising process was designed as a sprint. Khokhar booked 100 investor pitches across ten days. The opening stretch produced 47 straight rejections. Later, he and John Xing would remember staying on a friend's couch in New York, walking from office to office for two weeks and hoping someone would take the bet. The concentration eventually created its own momentum. They landed a meaningful first check, and Amigo acquired the other half of its founding pair.
Xing, Amigo's CTO, brought engineering experience in data and AI infrastructure. Khokhar describes the useful co-founder fit as a loop: the CEO should be able to sell what the CTO can build, and the CTO should be able to build what the CEO can sell. It is a tidy formulation, but the company's next two years made it practical. Amigo repeatedly sold a possibility, built it, watched how the market behaved and changed direction.
The initial AI-clone product grew. Khokhar found a dense pocket of coaches and expert-service operators inside a paid Slack community that cost roughly $20 per month. He did not enter with a broad campaign. He formed relationships in a room where prospective buyers were already trading notes. That small community led, by his account, to Amigo's first $1 million in annual recurring revenue.
By the end of 2024, Khokhar wrote that Amigo had grown from two founders to a team of eight, increased revenue 60-fold, outgrown two offices and raised a $6.3 million seed round co-led by General Catalyst and GSV Ventures. The company had also launched from stealth. Its Times Square appearance supplied a neat visual reversal: the founders who began the year walking between pitches were now looking up at their company on a screen.
The decision to erase good news
Traction created a more difficult question than failure would have. Khokhar could see a path for the SMB clone business to reach perhaps $10 million in annual revenue. He no longer believed the model could support the scale of company he intended to build. Continuing would have produced more customers, more code and more reasons to avoid a decision.
Then an enterprise agreement arrived that doubled Amigo's revenue. The contract pointed toward six-figure relationships and a narrower market. Khokhar chose it. Amigo refunded its existing customers, deliberately churned all of its revenue and focused on enterprise healthcare. In the telling he gave on A Product Market Fit Show, the company returned to its previous revenue high within two months and then grew tenfold, reaching $2 million in annual recurring revenue in under a year.
A market can validate a product without validating the company a founder wants to build. Khokhar treated the working SMB model as evidence, then separated that evidence from obligation.
This was less a technical pivot than a change in the burden of proof. A digital coach could answer a question. A clinical agent might be allowed to collect information, navigate a workflow, coordinate a handoff or take an approved action. Fluency would not be enough. The product needed boundaries, monitoring and a way to demonstrate how it behaved before deployment.
Trust becomes a product specification
Khokhar breaks trust into three pieces: control, alignment and observability. Control lets a clinical expert shape and constrain an agent. Alignment keeps behavior connected to an organization's changing requirements. Observability makes decisions visible in real time. The appeal of the framework is its plainness. Each word can become a product requirement, an evaluation and a reason to stop a launch.
Amigo starts each deployment with an “operable neighborhood,” a phrase Khokhar borrows from autonomous driving. An agent is assigned a defined set of situations where it can work. A trio consisting of an Amigo agent engineer, a customer's clinical expert and a product lead decides what good behavior looks like. Synthetic interactions then press on the boundary, especially with difficult or unusual scenarios. After launch, monitoring can feed new cases back into evaluation.
The architecture sits above foundation models rather than attempting to replace them. Different models can be routed to different tasks, while Amigo supplies memory, reasoning, behavior controls and orchestration. The practical division is deliberate: customers define clinical logic and experience; Amigo handles the systems that make those choices executable and measurable.
In March 2026, Amigo announced an $11 million Series A led by Madrona with participation from Optum Ventures, bringing reported funding to $17 million. The company said its agents had completed more than three million autonomous patient encounters during the prior six months. It also reported zero safety incidents in that period and a 100 percent pass rate in its pre-deployment safety evaluations. Those are company-reported measures, and the distinction matters in a story about verification.
The verification layer
Khokhar's long view reaches beyond any single workflow. “We believe we're moving from a human-based economy to an agent-based one.” If agents are going to take on consequential work, he argues, they will need something analogous to credentials: evidence that they perform reliably inside a specific scope, plus continuous observation as that scope expands.
The idea loops back to his earliest method. Before code, Khokhar used money to test whether demand was real. Before deployment, Amigo uses simulations and scorecards to test whether behavior is ready. In both cases, the interesting move is to replace a flattering story with a condition that can fail.
That habit also explains the hard reset. Khokhar did not treat revenue as a permanent verdict. He treated it as one reading from the market, useful but incomplete. The result is a founder story built around subtraction: ignore praise without payment, survive the noes, leave a channel that has reached its ceiling and narrow the product until the standard becomes legible.
Amigo now has the financing, customers and public claims that bring a different kind of scrutiny. Khokhar's task is no longer to show that someone will pay for the possibility. It is to make the company's definition of trust survive contact with scale. The bet is that verification will become its own layer of the AI economy. The obligation is to keep proving it.