The founder of Jotform, who started the company two decades ago when AI still sounded like a technology from the future, has laid out five things he would do differently if he launched it today. His argument is not that founders should adopt AI tools faster. It is that the company itself — how work moves through it, who it hires, how fast it ships — should be assembled around human-AI collaboration before the first employee arrives. He cites a recently published Anthropic guide to building a startup around AI as evidence that the barrier to starting has already fallen.
The analogy he opens with is hardware. A compact Nokia from the early 2000s and an iPhone 17 are not the same product at different stages of maturity; the internet era forced handset makers to rethink what a phone was for. His claim is that AI puts every executive in the same position, and that the productive question is not which tools to bolt on but what you would build if you started over. That reframing, he argues, moves a leader from staged adoption of new technology to reassembling systems from nothing.
The autobiography is there to make the point land. He describes his first years as trial and error: sole founder, self-funded, a small project built in college, no real experience managing other people, and a salaried job he kept for a while after launch. Roughly a year in, when he could afford help, the answers to millions of small questions about running a business — onboarding, training, benefits administration — started arriving one at a time. Anthropic's guide makes the counterpoint: AI has flattened the ground for anyone who wants to start a company or ship a product, and it has compressed the beginning. A founder today can put those same questions to ChatGPT or Claude and get answers immediately, along with market and competitor research, drafts and edits of documents for stakeholders, and an interlocutor who will argue with an idea without an interest in it.
The second recommendation is the sharpest. Design workflows, not job descriptions. Designing roles, he writes, is like adding lanes to a highway; designing processes is like building a rail network. The first raises throughput gradually, the second changes how work travels. AI-native startups do not begin by hiring people to do tasks X, Y and Z; they build systems that carry repeating workflows, and then hire people who can adapt and grow alongside those systems as the company scales.
Every organisation has the same logistical residue: routine HR questions, onboarding, process documentation, internal requests. None of it requires constant human supervision. An agent trained on a company's internal policies can handle recurring questions about pay or benefits, and the workflow holds whether the company has five employees or 500. Build that in from the start, and the operational load does not grow with headcount.
The third point follows from the second. Five years ago companies still prioritised narrow specialists; the AI era has shifted hiring toward people who work across business functions and direct the tools that absorb the routine. Founders should look for breadth of interest, curiosity about new tools, and the communication skills needed to instruct those tools and interpret what they return. In exchange, employees need training, time to experiment, and guidance. At Jotform, cross-functional teams expose people to how AI is used elsewhere in the business, and weekly demo days require staff to share recent experiments, the failures included; a tactic one group tried often reaches a team working on something unrelated. The company also rotates people into short AI initiatives outside their normal remit.
The fourth is about tempo. The sequence from idea to launch has not changed — generate, prototype, test, ship — but the pace has. Experimentation and user feedback are still required; what is new is that a small team with constrained resources can run the same process that once belonged to global companies with large budgets. The consequence he cares about is optionality: a pivot stops being a survive-or-lose moment and becomes an experiment that can be corrected.
The fifth reverses direction. Soft skills — now sometimes called power skills — are appreciating: empathy, communication, leadership, judgement. With school-age children building websites by vibe-coding, what separates professionals is what AI cannot imitate. The prescription is mentorship, empathy exercises, training in active listening, and an explicit instruction to automate as much as possible in order to free time for relationships and decisions.
The five points do not sit comfortably together, and the essay does not acknowledge the seam. Point two says stop hiring for tasks and build systems instead; point three says hire broad, curious generalists and then fund their training, their experiment time and their rotations. Both can be true, but the second spends what the first saves, and nothing in the piece explains how a five-person company pays for weekly demo days. Every concrete example comes from a business with twenty years of revenue behind it. This reads less like a founding playbook than like an incumbent's operating practices offered to people who have neither the headcount nor the slack to run them.
There is a second thing the essay is quiet about. The tasks nominated for automation — answering routine HR questions, onboarding, documenting processes, handling internal requests — are precisely the entry-level operations work through which people have historically learned how a company actually functions. The author says so himself: he learned the business by accumulating answers to millions of small questions. If an agent now answers them, the founder is spared the grind, and so is everyone who would have been formed by it. The essay names no replacement path.
The reliance on Anthropic's guide is worth a flag of its own. A model vendor's manual for founding companies on models is, among other things, a sales document, which does not make its central claim wrong but does leave its most interesting edge unpressed. If every founder gets the same instant answers from the same two or three models, those answers stop conferring advantage. Levelling the ground removes other people's hills and yours.
The strongest sentence in the piece is its last one: as AI takes over more of the execution, human value will be set by how well people think, relate and decide. The mechanics recommended to get there push the other way. Automate the rote, hire for breadth, compress the loop, and the founder ends up exercising judgement with less direct contact with the business than any previous cohort had. AI-native companies will be run by people who learned the job by watching a system do it.