For most of my career I’ve heard the same line. Building a software company is a marathon, not a sprint. Pace yourself. Compound slowly. Endurance wins. I think that advice is now wrong.
I don’t mean that in the lazy sense of “everything is different in the AI era.” I mean it in a specific way: the dominant variable in software companies has shifted from stamina to velocity, and most leaders haven’t updated their operating model to match.
I have worked with a lot of companies, and what I’m noticing is that products which took years to build and convinced investors and customers they had a defensible position can now be reproduced by a competent person with Claude in a weekend, and not a watered-down version either, but a real one. Moats that felt durable in 2023 have evaporated by 2026, and the only question left for the founders is how fast they can pivot.
That story isn’t isolated. It’s the through-line of what I’m seeing inside my own company, inside the companies I have advised, and across the public moves of operators I respect. The companies that will matter in five years aren’t the ones with the best plan, they’re the ones with the lowest latency between learning something new and acting on it.
This piece is an attempt to put that into a mental model I can actually use, and that I think other operators should adopt before their planning cycle bites them.
What I’m seeing
Three vantage points are pointing me in the same direction: my own company, the companies I work with, and the public moves of operators I respect.
Inside my own company
At TicketSwap, where I’m CPTO, the pace of shipping has changed in a way that’s hard to convey to anyone who hasn’t lived through it. I’m not going to throw out a multiplier, because the honest answer is that the variance is high and the gains are unevenly distributed across the org. But the trend is clear: things that would have been a quarter of work are now landing in weeks, and things that would have been weeks land in days. That has consequences I didn’t fully appreciate twelve months ago.
The first is that our strategy cadence has compressed. We used to run quarterly planning that genuinely held up for the quarter. Now the world moves enough between Q1 and Q2 that the Q1 plan needs a real rewrite, not a refresh. Annual roadmaps have become an artifact more than a tool. The actual unit of strategic thinking is closer to six weeks.
The second is that our roles are blurring. The clean lines between PM, engineer, and designer made sense in a world where the bottleneck was building. The bottleneck now is deciding what to build, validating it, and routing the next decision. The best people on my team are increasingly the ones who can move across that whole loop. We’re still hiring engineers, because our product surface is wider than the capacity we have to cover it. But the engineer I’m hiring now is a different profile than the one I was hiring two years ago: AI-native, faster to learn new domains, more comfortable owning a piece end-to-end.
The third is that the competitive picture has shifted. The threat is no longer the incumbent I used to track. It’s the AI-native challenger that didn’t exist 18 months ago and is now shipping something credible in our space. They don’t have our brand, our liquidity, or our trust, but they have an advantage I didn’t have to think about before: nothing weighing them down. No legacy code, no legacy assumptions, no legacy plan.
One example, up close
I won’t name the company. The lesson is general anyway. Their product was genuinely good, the kind of thing a small team of engineers and applied scientists had built up over years, with real IP, real customer love, and a defensible position in a category most people couldn’t have built in.
Then a couple of LLM releases happened, and the category collapsed underneath them. Not because Claude or ChatGPT replaced their product directly, but because the workflow their customers were paying for could now be done in a chat window with a model that costs orders of magnitude less.
What I’ve taken from watching their last twelve months up close is that AI’s commoditization runs through the underlying value those products were converting into revenue, not through the products themselves. When the value is “we save you 40 hours a month doing X,” and X becomes a €20-a-month subscription that any model handles natively, the product gets structurally repriced, not just partially devalued.
Their leadership conversations went from “how do we accelerate growth” to “how do we pivot in the next two quarters, not the next two years” in the space of one cycle. The hardest part, by a lot, wasn’t picking the new direction but accepting that the old one was gone. Boards and founders tend to be slower than markets to admit that.
Public signals
You can see the same shift externally if you look.
Shopify sent an internal memo in 2025 requiring teams to prove AI can’t do a job before they’re allowed to hire for it. The line that stood out to me: their revenue has grown around 21% a year while headcount went from 11,600 to roughly 8,100. They’ve turned operating leverage into an explicit cultural question.
Klarna moved aggressively in 2024, cutting around 700 customer service roles and routing them to an AI agent. In 2025 they walked some of that back, conceding they’d gone too far on quality. The interesting part isn’t the reversal but that even after correcting, their AI still handles roughly 75% of chats. The new baseline is permanent, even after the overshoot.
Duolingo announced an “AI-first” stance in 2025, gradually replacing contractors and shipping 148 AI-generated courses essentially overnight. They took heat for it and did it anyway.
Reasonable people will read these stories differently. I read them as a single signal: the operating model is being rewritten in public, and the leaders writing it are willing to take reputational hits to stay ahead of the curve. They’ve decided staying ahead is worth more than being liked.
The mental model
Two ideas. Together they’re what I’m trying to operate by.
1. Strategy is continuous, not episodic
The annual plan is a hangover from a world where the cost of changing direction was high and the rate of change in the market was low. Both have inverted: the cost of changing direction has dropped, since you can re-platform a feature in a sprint that used to take a quarter, and the rate of change in the market has jumped.
If your strategy is a document you write in November and live by until next November, your planning rhythm is misaligned with the world you’re operating in. The fix isn’t to throw out planning but to treat strategy as a question you re-ask every six weeks, with real authority to change direction, not a ceremonial review where the answer is always “still the plan.”
Ravi Mehta, who’s been on Lenny’s Podcast a few times, makes a useful distinction between velocity and latency. Velocity is how much you can ship; latency is how fast you can go from an idea to testing it. Big companies have velocity, startups win on latency. In an AI-accelerated market, latency compounds. Every cycle of “we saw something, we tested it, we acted” you complete while a competitor is still in their quarterly review widens the gap.
The implication for operators: redesign your decision cadence before you redesign anything else. If your strategy reviews happen quarterly and your competitors’ happen weekly, you’ve already lost on latency, regardless of how good your strategy is.
2. Velocity is the moat
Brian Balfour, co-founder of Reforge (which I can highly recommend!), made the point that “the window you have to get to escape velocity has shrunk.” Incumbents can copy faster, and distribution is more contested. What buys you durable advantage is no longer what you’ve built, since anyone can build it, but how fast you can change what you’re building.
This isn’t a new idea in spirit. Jeff at Ramp built a whole culture around the word “velocity.” Stewart Butterfield pivoted Slack out of a games company because he refused to be precious about what he was working on. The new thing is that the bar has dropped enough that this needs to be every operator’s default, not the unusual move of an unusually decisive founder.
A few practical consequences I’ve adopted:
Build for optionality instead of optimization. Stop tuning the current plan to perfection. Build the org so that switching directions is cheap. That means modular product surfaces, generalist hires, and short planning horizons.
Make pivots boring. If your team treats a pivot as a crisis, you’ll do it too late. Most of the companies I’ve watched wait too long do so because the cultural meaning of “we’re changing direction” is loaded. Strip the drama and make changing direction a normal thing that happens because the data changed. I’ll be honest, we still have room for improvement here, but it’s a deliberate direction I’m taking.
Treat your moat as a hypothesis. Every six months, ask: if a smart team with Claude started today, how long until they’re in our space? If the answer is less than a year, you don’t have a moat, you have a head start, and those are very different things requiring very different strategies.
What this changes on Monday
If any of the above resonates, here’s the diagnostic I’d run:
How long is your shortest possible loop from “we noticed something” to “we changed what we’re doing”? If the answer is measured in quarters, that’s your bottleneck, not your product or your headcount.
What does your annual plan assume about the AI landscape? If the answer is “roughly the same as today,” your plan is already wrong. Rewrite the parts that depend on those assumptions as scenarios, not as load-bearing facts.
What part of your current value proposition would survive if a frontier model could do the underlying job for €20 a month? Answer that honestly, and treat whatever wouldn’t survive as your sprint priority.
Are you hiring for velocity or for stamina? Most companies still run their hiring loop for the latter. The profile that wins now is closer to a generalist who can learn a new domain in a week than a specialist who’s been doing one thing for ten years.
None of this is comfortable. The marathon mindset was comforting for a reason: it absolved you from re-deciding the thing every week. Endurance and consistency felt like virtues, and they were, in a different market.
The market we’re in now rewards a different virtue: the willingness to admit, on short timeframes, that the thing you committed to is no longer the right thing, and to move before the evidence is overwhelming.
It’s a sprint now, so run it that way.
Super interesting read, very relevant for me right now as well.
The uneven distribution rings true, would be curious to discuss some of the higher gains applications within Ticketswap!