Danny Castonguay has a question for the boss

Danny Castonguay begins the lesson with a stuffed toy. Give it to a child. Let her name it and decide where it sleeps. Take it away ten minutes later. Her possessions are back where they started; her feelings have acquired a grievance. Now replace the toy with a department, a budget and a title. The meeting has become rather more expensive.

In September 2026, the co-founder and CEO of bld.ai used that thought experiment in a letter to his daughter Ada. He imagined a manager who discovers that a smaller team can serve customers better at half the cost. The evidence is convincing. Unfortunately, the improvement threatens the manager’s own position. More studies are requested. More approvals become necessary. A sensible proposal develops a mysterious inability to leave the room.

The letter ends by asking his daughter to remember the toy if she ever gains the power to make others pay to preserve her status. It is a revealing concern for someone whose business involves helping organizations change. Castonguay is interested in the machinery beneath the respectable explanation: incentives, attachment, the distance between what serves the customer and what preserves the chair.

The classroom came before the platform

His route into that machinery began with engineering. A Montreal native, he studied at McGill, worked at McKinsey and went to MIT Sloan for an MBA. Along the way came AntPortal, software for documenting research and development work. The managerial ambition was already quite explicit. “I’ve known for a long time that I wanted to be the CEO of my own company,” he said in 2013.

At MIT, he became a Legatum Fellow. In 2012, he went to the University of the Philippines Diliman through MIT’s Accelerating Information Technology Innovation program. Teaching entrepreneurship brought him into contact with developers whose skills he thought deserved local opportunities. In an interview at a developer event in Makati, he argued for building software companies in the Philippines and keeping the talent there.

The work also became a venture. He helped build Kalibrr as co-founder and chief technology officer. During the fall of 2012, he recruited more than a dozen MIT students to help with fundraising, market research and a beta platform. His assignment included improving software engineering processes so the team could release more quickly while protecting product quality. The connection between teaching people and organizing their work was becoming practical.

There is an earlier classroom souvenir. As a software engineering teaching assistant at McGill, he received an evaluation calling him a monkey. He remembers laughing. In a later reflection, he wondered whether his slow walkthroughs of examples had contributed. Other students had described him as patient. He acknowledges that he has also failed at patience. A teacher who can keep the unflattering review has at least retained some useful course material.

The company that had to inspect itself

After another period at McKinsey, Castonguay built bld.ai. His résumé describes work across sectors including energy, logistics, retail and banking, combining software engineering with business problems. His concern about wasted talent acquired a corresponding concern about wasted capital. In a 2025 essay, he criticized the comfortable spending habits of large enterprises and generously funded startups alike. Having money, in his telling, does little to guarantee a worthwhile use for it.

At bld.ai, the coordination of professional work became a software problem of its own. The company’s internal platform, Flow, helped run the business. In his May 2023 retrospective, Castonguay reported 500 members, $14 million in revenue and $3 million in profit for the previous year. He also said 95 percent of members worked as contractors. These were figures for 2022, describing a network whose organization mattered as much as its size.

The interesting turn came after the figures. Sales relied on too few people. An internal platform had users who were, in effect, captive. A growing network could no longer rely on everyone knowing everyone else. He wanted broader adoption of the tools and more systematic checks on delivery. Revenue had supplied evidence of demand; it had also financed a new set of questions.

That habit of moving from presentation to interrogation runs through his advice on selling. He describes himself as someone who prefers tinkering to a sales identity. His approach asks what a buyer actually faces, which complication matters and whose incentives govern the decision. One example involves a corporate vice president; another involves a four-year-old. Both require understanding the person before proposing the solution. Neither is guaranteed to approve the budget.

Eight apps, and a bill for understanding

Castonguay enjoys trying the tools himself. In April 2025, he described creating more than eight apps over twenty days, often in hours. His examples included a pirate-themed browser game, OOPSDEAD. The process began with a voice note, moved through a structured project brief and a generated prototype, and used additional services for AI features. He also described a point at which agents became frustrating and he wanted a capable human team.

That account makes his later warning about AI programming more interesting. In October 2025, he wrote about theory debt: the understanding a person fails to acquire when software arrives before a mental model of how it works. A generated feature can seem almost finished. Trouble appears when someone needs to change it and discovers that dependencies, assumptions and design decisions remain obscure.

You can build quickly and still owe yourself an explanation.

The tension in Castonguay’s writing on AI-assisted development

He connects the problem to computer scientist Peter Naur’s account of programming as theory building. Knowing the system means knowing how it relates to the world and why its choices make sense. Producing code does not automatically produce that knowledge. The practical consequence is familiar: a small change becomes an expedition. His enthusiasm for prototypes survives the warning. So does the obligation to understand what has been made.

A month earlier, he had been at the ICPC World Finals in Baku, discussing learning with competitive programming coach Fegla. His recommendations paired AI assistance with deliberate practice. Easier problems build fluency; harder ones expose gaps. For hiring, he favored observing problem-solving under constraints. The tools can help, but people still need to formulate the problem and own consequential decisions. The classroom had followed him to a programming contest.

Please bring a better paragraph

His management writing frequently asks the reader to supply something concrete. In a November 2025 feedback matrix, he separates agreement from disagreement, and both from the effort put into a response. A friendly suggestion can leave all the labor with its recipient. A criticism accompanied by a rewritten passage or an alternative design brings something the recipient can use.

It is an appealing standard for anyone who has received a document decorated with helpful questions. Questions are inexpensive to manufacture; answering them has a labor cost. Castonguay gives thoughtful disagreement value, then asks whether the critic can improve the work. The framework makes participation visible. One person has brought an opinion. Another has brought an opinion and rolled up a sleeve.

His instructions for email pursue the same economy. Put the decision or request first. Distinguish what happened from what you think it means. Name the next action. He allows AI drafting and objects to generic language. A sentence should belong to the particular situation. This is directness expressed as a service to the reader: fewer ceremonial introductions, a clearer account of who must do what.

An education with customers

The lessons travel home. In July 2026, Castonguay wrote about teaching children to create things other people want, then to consider whether those things deserve to be made. He described his eldest daughter’s work moving across drawing, craft materials, sewing, photography and short videos. Each medium introduced a skill or a different way of reaching an audience. The projects gave independent choices somewhere to become visible.

He distinguishes practicing a technique from deciding what to create. Both matter, but they ask different things of a learner. A piece can be reproduced accurately without anyone deciding that it should exist. A venture introduces the additional inconvenience of other people’s preferences. That is a useful bridge between a child making something and a founder discovering that a perfectly functioning product attracts little interest.

His August 2026 essay offers another piece of preparation. Before a difficult interview or contest, he deliberately accepts the possibility of finishing last. He applies that idea to founders and to his children’s jiu-jitsu competitions. Once failure has been admitted, he finds more room for curiosity and learning. It is an unusual prelude to competing: first give up the imaginary guarantee, then get on with the attempt.

Which brings the story back to Ada’s stuffed toy. Across software, classrooms and companies, Castonguay keeps returning to the decisions people can make once they understand what they are doing and what they are protecting. He has wanted the CEO’s chair for a long time. His letter asks what happens when keeping the chair becomes the purpose. For a founder, that question belongs within reach of the desk.