Daniel Erickson built a company to help other companies listen. Then the customers began asking for something he had not set out to sell. Large businesses were turning up at a tool meant for young startups. They already had products people wanted. What they lacked was a practical way to understand everything those people were saying. For a founder interested in feedback, this was an awkwardly useful piece of feedback.
Viable began in 2020 with a narrower ambition: make the search for product-market fit less mysterious. It became an early commercial application of GPT-3, translating piles of customer comments into answers that colleagues could read without a lesson in databases. The company closed in 2024. Erickson moved to Airtable and then Vercel. The question running through those changes has been remarkably consistent: how do you get useful information into the hands of someone who can do something with it?
Two brothers, two ways into a product
Erickson grew up in Oregon with his identical twin, Jeffrey. They started their first company at 16. Daniel became an engineer; Jeff became a designer. By 2006, they were running an application development agency in Portland, building prototypes for early-stage founders. The brothers shared an interest in products while approaching them from different sides of the screen.
The arrangement gave Daniel repeated encounters with the beginning of a business: a founder, an idea, the first software, the effort to turn a proposition into something another person could use. Eventually he wanted to stay with one product beyond its prototype. He moved into the Bay Area startup world, joining Yammer as an early engineer and bringing Jeff into its design work.
His route included studying computer science at Southern Oregon University from 2006 to 2008. He left without a degree, describing his professional skills as largely self-taught. His LinkedIn education entry lists an extracurricular activity with some economy: “Dropping out.” It is a small joke at the expense of the standard career form, whose boxes were evidently designed for more cooperative students.
At Yammer, he saw a company growing with demand behind it. At Getable, where he became chief technology officer, finding that demand proved harder. He later led engineering at Eaze. The contrast between those experiences kept product-market fit on his mind. Having worked inside businesses with different relationships to their markets, he wanted a method that founders could use repeatedly.
A measuring instrument meets its own test
The initial idea drew on Superhuman’s approach to measuring product-market fit. David Sacks shared a link to Rahul Vohra’s account of the process, and Erickson saw work that software might take off a founder’s desk. Surveys could run continuously. Responses could be grouped. Product teams could use the results to decide which strengths to deepen and which frustrations to fix.
He and Jeff built Viable Fit around that opportunity. The irony was available almost immediately: a business selling help with product-market fit still needed to find its own. Early-stage startups had the problem, but they were a limited market for a paid solution. Meanwhile, larger companies were using the tool for reasons that did not fit the original pitch.
Those customers had feedback scattered across support systems, reviews, surveys and conversations. The useful part of Viable was its ability to make sense of the text. The founders followed that demand toward broader customer feedback analysis. Their customers had effectively edited the business plan. Accepting the edit required giving up the comfort of the original brief.
“I build systems that help teams do their best work.”
Daniel Erickson
A month spent asking the machine questions
Erickson received early access to GPT-3 and spent about a month experimenting with the model and Viable’s data. He came away convinced that a question-and-answer interface was possible. A customer could ask something in ordinary language and receive an answer grounded in the feedback, expressed in ordinary language too. It offered a shorter route between curiosity and a decision.
In March 2021, OpenAI featured Viable among applications using GPT-3. Its demonstration centered on a familiar commercial nuisance: a frustrating checkout. Rather than return a cloud of words, the system could summarize specific complaints about loading speed, address editing and payment options. The ambition was to preserve the reason for the complaint, rather than reduce the customer to a red mark on a sentiment chart.
The engineering involved choices beyond the language model. Erickson described using more than a dozen additional internal and proprietary models, and treating cost as a product constraint. A service that answered questions brilliantly at an unsustainable price would present its founders with another, less entertaining question. The work involved deciding which capabilities to build and which to buy.
He later set out a distinction between products built around prompts and products built around data pipelines. The former package an interaction with a model inside an interface. The latter depend on access to information and a sequence of processing steps. He argued that the underlying data could make the second kind harder for a general-purpose model to reproduce.
That was also a description of the business he was trying to build. Viable's value depended on the customer's actual material and the work of organizing it. A fluent answer alone could not supply the missing evidence. The interface might feel conversational, but the proposed advantage lay further back, in what the system could read and how it handled what it found. It is a useful distinction for understanding why Erickson spent so much time discussing the less photogenic parts of an AI product.
The T-shirts had a shipping problem
One of Erickson’s examples involved an ecommerce business selling custom-printed items. Customers complained about damaged T-shirts. An analysis of support tickets pointed toward shipping rather than manufacturing. The company changed shipping providers. In his account, complaints about quality after shipping fell by 50% the next month. The example is useful precisely because the answer was so unglamorous. Nobody needed to reinvent the T-shirt.
He also described the labor preceding such an answer: a product colleague collecting exports, putting comments into a spreadsheet, tagging them into broad categories and writing a summary for leadership. Every handoff could remove detail. The final paragraph might be easy to read while being too general to tell anyone what to change.
His criticism of conventional feedback handling was specific. A topic such as checkout tells a team where to look; it does not necessarily identify what went wrong. He wanted the complaint, the circumstances and the possible response to survive the journey into a report. The next round of feedback could then help establish whether the intervention had worked.
The team is part of the system
Erickson applies a similar interest in communication to management. His public introduction organizes the job around people, process and purpose. Listening and clarity matter because a team needs to work together. He has little patience for the idea that individual brilliance excuses making everyone else’s work unpleasant.
In his writing on hiring, he names five qualities: tenacity, decisiveness, empathy, flexibility and candor. The combination is revealing. Persistence sits alongside a willingness to change direction; frankness has to coexist with an ability to understand another person’s view. His suggested interview questions ask candidates for examples of their behavior under pressure, rather than an agreeable recital of values.
He also writes about speed with a qualification that matters. Reversible decisions deserve a different pace from decisions whose consequences are difficult to undo. Move quickly where a correction is cheap; spend more thought where it is expensive. The attraction is a faster cycle of learning, rather than the theatrical pleasure of appearing decisive in every meeting.
His technical writing offers a related clue. In “Atomic Product Development,” he argued for bringing design and development closer through reusable modules, and introduced Atomify as a tool to help bundle them. The concern was practical: reduce the repeated handoffs and accumulated mess that make a product harder to change. Better arrangements could spare people work they should not have had to repeat.
The founder changes the job title
In July 2024, Erickson announced that he had decided to shut Viable down after four and a half years. He thanked the team, customers and investors, then described his next role at Airtable. There, his question was how AI could help people with less technical experience create internal and external tools. His announcement accompanied the launch of Airtable’s Cobuilder.
That move carried a recognizable idea into a different setting. At Viable, ordinary language had been a way to ask questions of customer feedback. At Airtable, it could describe an application someone wanted to make. Both reduced the distance between a person’s intention and the technical work needed to express it.
By July 2025 he was publicly discussing his decision to join Vercel, singling out the engineering taste of colleagues including Lars Grammel. His current public X biography identifies him as engineering director for Agent Stack, listing the AI SDK, AI Gateway and Workflow DevKit. The founder of a feedback product was now working in the tools used to build AI software.
On his personal site, Erickson describes systems thinking and narrative as complementary interests: one helps him map how things work, the other helps him understand human motivation. That pairing gives his career its coherence. Software has to process information; people have to recognize why the result matters. A customer’s words, a designer’s intention, a colleague’s objection: each can disappear inside a badly arranged organization. Erickson has kept returning to the arrangement, trying to make it easier for the useful part to get through.