Overheard September 2026: AI Founders on Work, Management & Culture
What founders are saying about the way AI is changing work inside their own companies.
Image created by the Superhumxn team.
Much of the narrative around AI and work is still about what might happen next. So what does work look like inside companies building and adopting AI right now?
This month’s Overheard brings together some of the more revealing comments founders have made about how their companies are being run, from the way engineers work to how employees are assessed.
What can I expect from a subscription?
Real-world strategies for navigating the future of work.
AI AGENTS AT WORK
Everything you need to know about how AI agents are changing teams and ways of working.
SUPERHUMXN IRL
Real insights from investors and operators.
MEMOS ON PEOPLE, CULTURE & AI AT WORK
The latest use cases and strategies for change and transformation.
Tobias Lutke, Shopify
"We call those 'slop grenades' that people toss at each other."
Shopify CEO Tobias Lutke has been pushing employees to use AI for some time. Last year, he told staff that using it was becoming a baseline expectation at the company. Teams asking for more people would first have to explain why they couldn't get what they needed from AI.
Since then, Lutke has noticed a problem. He told The Knowledge Project that employees can generate an email or piece of code, barely check it and send it to a colleague. The recipient then has to work out what is wrong with it.
"Slop grenade" describes the problem rather well. AI has saved the sender time by moving some of their work onto somebody else.
Ivan Burazin, Daytona
"We've found that people who manage people and understand how to communicate what they want, inputs versus outputs, they're much better at running agents."
Nobody at Daytona writes code anymore, according to CEO Ivan Burazin. The company has 16 engineers and each runs around five AI agents.
Burazin told Business Insider that engineers who had previously managed people adapted well to working this way. They were used to explaining what they wanted and giving someone else enough information to do it. Some strong individual contributors found that harder because they were used to carrying more of the context themselves.
The engineers are still responsible for producing software. They're spending less of their time writing it and more of their time directing the agents that do.
Matteo Franceschetti, Eight Sleep
"Our engineers stopped coding around a year ago."
Eight Sleep co-founder Matteo Franceschetti told 20VC that its engineers now work through AI agents. He said the company had made similar changes elsewhere, including replacing work previously handled by an email marketing operation with bots built in three days.
Franceschetti is still particular about who joins the company. He said he would be wary of a candidate deciding between Eight Sleep and Anthropic because he wants people who specifically want to work at Eight Sleep.
He is changing the work quickly without lowering his expectations of the people doing it. Someone joining Eight Sleep may spend much of their time directing AI, but Franceschetti still wants them to have a clear reason for choosing his company.
Image created by the Superhumxn team.
Liang Wenfeng, DeepSeek
"No one manages them. No KPIs."
DeepSeek founder Liang Wenfeng has said that employees generally don't work overtime. Researchers are also given substantial amounts of time without prescribed work because he believes research needs room to think.
There are no monthly KPIs. Liang has also said that the shared vision holding the company together hasn't been formally written down.
DeepSeek has kept that approach while growing and competing for AI researchers. Rather than adding more targets as the company gets bigger, Liang has deliberately left researchers with time that isn't assigned to anything.
Marc Lore, Wonder
"AI basically calculates whether or not you should be promoted."
Every six months, Wonder employees are rated by at least a dozen colleagues. Their scores and written feedback go into an AI system alongside the company's "value above replacement" measure.
CEO Marc Lore told Fortune that the system recommends whether an employee should be promoted and how long they should remain at their current level. Managers can override the recommendation, although Lore says they do so less often than they used to.
Lore calls the process "very objective". Some of the information going into it is still based on what colleagues think of one another. AI can apply the same process each time, but that doesn't make a colleague's assessment objective.
Rune Kvist, AIUC
"The surprising thing about AI is that it becomes harder to adopt and harder to control as AI gets smarter, not easier."
Rune Kvist was an early employee at Anthropic before starting Artificial Intelligence Underwriting Company with former METR COO Rajiv Dattani. AIUC announced a $40 million Series A in September and audits AI agents before companies put them to work.
Kvist told TechCrunch that more capable agents can be harder for companies to adopt. An agent might be perfectly capable of doing a job, but doing it may require access to systems or information the company isn't comfortable handing over.
AIUC puts agents through thousands of tests before a human verifies the audit. Companies want the work an agent can do without having to guess what it might do once it has access to the rest of the business.
Lucy Guo, Passes
"I worked a 26-hour day because I refused to go to sleep."
Lucy Guo says AI has made it harder for her to stop working. The Passes CEO and Scale AI co-founder told Fortune Daily that having agents available means there is always something else she can ask them to do.
Guo described herself as the first person in the office and said founders set the pace for the people around them. Her 26-hour working day is an extreme example of what can happen when AI removes some of the natural stopping points from work. Finishing a task faster doesn't create free time if you immediately use that time to start another one.
Overheard
There isn’t one version of the AI-enabled company emerging from these conversations. The technology may be similar, but founders are making their own decisions about where it belongs.
Daytona’s experience is a good example. Its engineers haven’t disappeared because agents can write the code. Their work has shifted towards giving instructions and checking what comes back, and people with management experience have adapted well to that change.
Wonder is using AI somewhere more consequential. Its system now has a role in decisions about employees’ careers. Shopify has run into a different problem, with AI making it easier for employees to pass poorly checked work to one another.
These are much more useful signals than another prediction about how AI might change work in five years. They show what happens once companies stop experimenting with the technology and start building it into the way they operate.
