True North by True Ventures: Founders are budgeting tokens like salaries
This issue is about what happens when building gets faster and cheaper. Jon Callaghan argues that AI won't be winner take all, it will be many take a lot, and you can see why in how teams are working now. Founders are budgeting AI the way they budget people. Inside True, we replaced software we'd used for a decade in a matter of days. And across the portfolio, small teams are doing work that once took whole companies: designing medicines with AI, putting radar satellites in orbit, and building a ring you can whisper your ideas into.
SIGNAL
The best chip for a frozen model is the model itself. Trained models can now be compiled straight into logic gates, a fixed circuit with no matrix math, printed as a custom chip. Every update means a new chip, so it suits models that stay put. Our prediction: once code can reach custom silicon, AI rewrites everything above the foundry.
Nobody has a benchmark for healthy token spend. At our October Founder Conversation, founders said they budget tokens like salaries: a $200K engineer works out to about $17K a month. The clearest test is tying spend to valuable work, and you can't judge an agent's output until you've defined good work.
AI services firms have to earn their software multiple. They start valued like services businesses, pricing a fixed outcome instead of selling seats to experts who bill hours, and keeping the margin AI creates. The proof is gross margin holding above 65% and more work shipping with no human touching it.
INSIDE TRUE
Two of this month's rocketship companies came to us through founders. In summer 2020, Fauna Bio's Ashley Zehnder introduced us to Viswa Colluru, and we led Enveda’s seed weeks later. A year after that, Viswa introduced us to Glen Gowers and Oliver Vince, and we led Basecamp Research’s second round. "Ashley's introduction did not make the investment decision for us," Rohit Sharma writes. What it gave us was "trust before there is a transaction." Rohit on how trust travels →
We replaced software we'd used for a decade in days. "We have weird workflows. So we rebuilt it," Mike Montano writes. Keerthi Nalabotu did the first pass of Truebase, and now an agent pulls the 20 similar companies we've seen over two decades, "why we passed, what we wanted to see and never did." The math on build vs. buy, he says, "is breaking." Custom software is "how we free up time to spend more time with extraordinary builders."Read more from Mike →
"AI will not be winner take all. It is many take a lot." While much of venture rushes into $100 million "seed" rounds, our co-founder Jon Callaghan sees "a market defined by abundance." "AI is the defining long-term innovation trend of our lifetime, and we will build and invest in it for decades to come," he writes. He's already seeing it at Polsia: "tremendous democratization of capitalist opportunity." The founders who win "will build big castles over many years."Read Jon's Scarcity vs. Abundance →
FOUNDERS BUILDING
ICEYE went from a tent on a Helsinki rooftop to the largest radar constellation on the planet. In 2018, Rafal Modrzewski and Pekka Laurila launched the first radar satellite under 100 kilograms. The rooftop story → Now governments across Europe are buying ICEYE fleets of their own, and with Nokia, ICEYE is building satellite networks countries can own. We think every country that wants to see for itself will turn to ICEYE. ICEYE and Nokia →
Polsia went from $0 to a $20M run rate in 28 weeks, and it's still speeding up. $0 to $1M took 12 weeks, $1M to $10M took 10, and $10M to $20M took 6. Now Ben Broca's company is about to pass $25M, and his next stop is $100M ARR. "The road from $25M to $100M is going to look very different," he writes.Ben tells the story →
Enveda has three drug candidates in human trials. Rohit Sharma led our seed investment in 2020, when Viswa Colluru's plan to use AI to read nature's uncharted chemistry was simple to describe and hard to execute. We've invested in multiple rounds since, and Enveda just raised $311M to push those trials forward and start more.Enveda's announcement →
Basecamp Research is designing medicines with AI. Glen Gowers and Oliver Vince set out to record the life science had never captured. Their EDEN models now learn from biological data gathered with partners in more than 30 countries, and a $140M Series C takes their first cell therapies toward human trials. Basecamp's announcement →
Stream made TIME's best inventions of the year. We backed Mina Fahmi and Kirak Hong in 2025 to build a new way to talk to AI: a ring you whisper into to capture notes and think out loud without pulling out your phone. Congrats to the Sandbar team.Stream on TIME's list →
TEAM TIME
Choose the partner you can call on a hard day. Puneet Agarwal met Duo's founders before they had an office, and we backed every round after. On TechCrunch's Build Mode, his advice: ask other founders how a partner showed up on their worst days. Can you bring them bad news, disagree, and talk straight about what's next? Puneet on Build Mode →
Your next builder might be a VP.
Mike Montano ran a 3,000-person engineering team at Twitter. Now he's building again, and he's found that all those years of judgment make him faster with AI. His advice: hire the managers, directors, and VPs who miss building. Some of your best hires might just be waiting for the chance. Go build →
"Only a person who risks is free."
In a quick street interview with Viraj Ala, Puneet shares the best advice he ever got, from his grandfather. "If you don't go and take a chance, you're never going to find the treasure at the end." Watch Puneet →