Barry McCardel and the trouble in the other tabs

Barry McCardel had good news for the boss. His team had returned from a customer visit with a new product installed, training completed, and users apparently converted. One woman could demonstrate the workflows. The rollout looked successful. Then came a question he had not prepared for: what other browser tabs did she have open?

He initially found the question intrusive and beside the point. On the next visit, he looked. The new product was there, as promised. So were a dozen tabs belonging to the old one. The customer had learned to perform enthusiasm for visiting software people while continuing to use the tool that helped her finish her work.

Once he asked about those tabs, the conversation changed. Their handsome interface loaded too slowly. A chart option his team had treated as optional mattered enormously to her. Other people on the floor were frustrated, too. The successful rollout had acquired a rather inconvenient audience review.

McCardel recounts the episode with himself as the person who missed the obvious. It is a useful introduction to the co-founder and CEO of Hex, the San Francisco-based analytics company. He likes software, certainly. He is also interested in the elaborate arrangements people make to survive it.

An airline inside a spreadsheet

Long before Hex, McCardel was working with data at Northwestern University, where his studies included network science. He then went into consulting, finding that Excel could be persuaded to do things far beyond the modest expectations suggested by its grid. In a later interview, he described this period as doing “unholy things in Excel.”

One example stayed with him. He worked with an airline whose flight schedule was optimized in a spreadsheet. Billions of dollars of decisions depended on a workbook whose logic he found both impressive and alarming. Python might have suited the task better. The people doing the work had Excel, and they had made it work.

That experience gave him a lasting respect for the spreadsheet. An empty cell asks very little of a newcomer; a skilled user can build something surprisingly elaborate in the same environment. McCardel wanted data tools to offer that combination of an easy entrance and room to grow. Removing every difficult capability would defeat the purpose.

In his early writing about code, he dwelled on the obstacles surrounding it: drivers, credentials, environments, dependencies. His analogy was email. Imagine needing to configure a local drafting environment before sending a message. People might decide that writing was the problem. Software has a talent for making its own paperwork look like the user’s inadequacy.

“unholy things in Excel”Barry McCardel, on his consulting years

He went shopping. Then he started a company.

A friend from college suggested Palantir. McCardel spent roughly five years there, working on data problems and meeting the two people who would become his co-founders, Caitlin Colgrove and Glen Takahashi. The relationships preceded the company. They had already seen one another work before agreeing to build something together.

After Palantir, another job brought a familiar frustration back into view. Organizations could invest in modern warehouses and other data infrastructure while analysts still passed work between disconnected tools. Queries lived in one place, notebooks in another, and finished charts made their way into documents as screenshots. The storage had advanced; the handoff remained clumsy.

McCardel began looking for software to buy. He asked friends on data teams how they handled the problem. Their makeshift answers failed to satisfy him. He has described himself as a reluctant founder: someone seeking a particular tool, eventually persuaded that building it was the available route. Hex began in late 2019.

The product brought analysis and sharing into a common workspace. People could work with SQL and Python, collaborate, and turn their work into interactive applications for colleagues. This addressed an unglamorous but consequential gap: a good answer has limited value when the next person cannot find it, use it, or ask a follow-up question.

Two customers, and a useful amount of trouble

McCardel’s advice about early customers is unusually numerical. The right number of design partners, he argues, is two. One risks producing software fitted to a single customer’s peculiarities. Too many can scatter a small team’s attention. Two provides a second test without turning product development into a committee tour.

He pairs this with an approach he calls commitment engineering. Start by finding people who recognize the problem. Ask whether they will spend time trying a rough solution. Return with improvements and ask for a larger commitment: a colleague in the next session, perhaps, or a real meeting conducted using the product.

The requests grow alongside the usefulness of the software. Payment comes after a series of exchanges that reveal whether anyone values the work enough to spend time or reputation on it. A pleasant conversation is easy to obtain. Persuading someone to put your unfinished application in front of their boss is a more demanding review.

The browser-tab story explains why this matters to him. People are often kind to the person who built the thing. They will learn the demonstration, say encouraging words, and quietly return to their old routine. His method tries to find behavior that survives after the visitors have gone home.

A bigger company, with the same awkward question

By October 2021, Hex had announced a $16 million Series A led by Redpoint. A $52 million Series B followed in March 2022, led by Andreessen Horowitz, with Snowflake and Databricks among the investors. Those names placed the young company alongside businesses building the infrastructure its users already relied on.

In May 2025, McCardel announced a $70 million Series C and reported that more than 1,500 teams worldwide used Hex. He also pointed to the arrival of the Hashboard team following an acquisition. The numbers described a company with a broader constituency than the few collaborators needed to test an early prototype.

Growth creates another version of his original problem. More employees bring more ways of working, and the urge to standardize becomes difficult to resist. McCardel has written about an internal debate over people moving between roles. Some had interviewed; others had rotated or demonstrated their suitability differently. The moves appeared to be working.

He asked what harm the variation had caused. The result was a brief document with guardrails. The episode captures his suspicion of procedures introduced merely because differences exist. He accepts that sales playbooks and other processes can help. He wants the proposed remedy attached to an actual difficulty before it becomes everyone’s new chore.

The AI builder who still writes his own argument

Hex’s move into AI introduced a new source of possibility and uncertainty. McCardel has described the first version of its Magic features as a month’s work by an engineer, a designer, and himself. The initial interaction was straightforward: supply a prompt and receive code. More elaborate capabilities followed.

He also tells a less tidy development story. The team spent time speeding up AI code edits, only for a faster underlying model to make the earlier approach adequate. It is a practical warning about building on rapidly changing technology. A problem can consume your engineers today and disappear in somebody else’s release tomorrow.

His outlook on analysts is more expansive than a simple replacement forecast. In a 2023 essay, he used Jevons’ paradox to argue that cheaper analysis could encourage more demand for it. Companies have questions they never get around to answering. Faster tools could change which questions become worth pursuing. He presents this as an argument about the future, with uncertainty intact.

By April 2026, he was also asking colleagues to preserve something slower: writing their own reasoning. At Hex, a lead proposing a hire writes a hiring thesis. McCardel says the exercise has changed his own decisions. On one occasion, after indicating he would proceed with a candidate, he sat down to explain the choice and discovered he could not justify it.

He acknowledges using AI for research, organizing notes, and challenging ideas. For consequential arguments, he wants to do the writing himself. The unfinished document has a function even when nobody reads it. It can expose a weak premise before that premise acquires a salary, a roadmap, or a meeting series.

Room for the people who do things differently

That preference also appears in his April 2026 memo, “Weirdos wanted.” He asks hiring teams to look for unusual interests, independent thought, and people who have gone down their own rabbit holes. His argument is that an organization working with AI needs people who contribute ideas beyond familiar patterns. It is his hiring philosophy, rather than a settled verdict on what models can do.

Nor does he treat his former employer as a complete blueprint. In his essay on forward deployed engineering, McCardel distinguishes Palantir’s costly, experimental field model from Hex’s approach. Hex sells one software product and does no custom development in the field. He wants close contact with customers while choosing a different business structure.

Outside Hex, he makes small investments in companies including Browserbase, LangChain, MotherDuck, Modal, and Railway. He also writes short, pointed essays about work: titles, design partners, variation, shipping. The recurring interest is how good intentions acquire machinery. A title can change incentives. A process can outlive its purpose. A polished interface can conceal the chart someone actually needs.

The customer with the old browser tabs offers a fitting last word on all of it. She was getting her work done, using the tools available, while the visitors congratulated themselves. McCardel learned more when he let her explain the inconvenience. Years later, with investors, AI agents, and a growing company in the picture, that remains a demanding instruction: look up from the thing you built.