Pieter Levels recently said the quiet part out loud: revenue and traffic are declining across indie hacker projects, his own included. He suspects the same is true for VC-backed products, but they don’t publish numbers, so nobody can check. His diagnosis is blunt. Big AI is cannibalizing everything that used to be an app. His tone, though, is calmer than the diagnosis. “Just times are changing, and we have to adapt.”
A few days later, Fireship covered it and made the stakes explicit: indie hackers may be the first type of developer to go extinct.
For about a decade, the playbook was stable. Learn to code, find a niche, ship a micro-SaaS, build in public, post the MRR screenshots after every failure, repeat until something sticks. It worked because code was expensive. Ideas were cheap, and execution was the moat, so the person willing to grind through the building had an edge over the ten people who had the same idea and never shipped it.
Execution got repriced
That equation has flipped. Execution now costs about $20 a month. Fireship framed it as a question: why pay $29 a month for a product when Claude Code can build a better version in twenty minutes? Two years ago that question was silly. Today I ship almost everything with Claude Code, and for a lot of the tools I used to pay for, I don’t have a good answer. Neither does Levels. He mentioned in the replies that he cancelled most of his SaaS subscriptions except the AI ones.
The replies under his post read like a support group. Jon Yongfook reported his Google referrals roughly halved in 2026, with signups sagging along with them. Even Tailwind is reportedly seeing sharp traffic drops, and Tailwind is about as far from a weekend micro-SaaS as an indie project gets.
The mechanism is a plain behavior change. Nobody googles “free invoice generator” and clicks the third result anymore. They ask an agent, and the agent doesn’t click ads or start free trials. The pipeline from search to signup that quietly funded a decade of indie software is being routed around.
The floor dropped
Fireship’s sharpest point was that AI raises the ceiling and lowers the floor at the same time. The ceiling part feels great when you’re the one shipping faster. The floor part is the problem: the field now includes everyone who can describe software in plain English, and their vibe-coded tool only needs one satisfied customer, themselves.
There’s a bitter footnote for the build-in-public crowd. Your public roadmap used to be marketing. Now it also works as a spec sheet that a stranger can paste into an agent the same afternoon.
What still compounds
Levels’ answer is distribution, and he has receipts on this one. He says he asked ChatGPT back in 2022 what would still matter once AI could code; it told him distribution, and he started building an audience seriously that year. Whatever you think of taking career advice from a chatbot, the timing worked out.
The playbook he and others in the thread converge on is not exotic. Show up on X every day with real numbers and real failures, and watch views instead of follower counts, because plenty of huge accounts get no reach in 2026. Reply fast and substantively to bigger accounts in your niche rather than waiting to be discovered. Move people onto an email list you own, because every feed you grow on is rented land. Make video, since personality still doesn’t compress into a prompt. Be useful in niche communities long before you pitch anything there. And increasingly, build the audience before the product, so you launch to people instead of at them.
I’d add one thing, because it’s where I’ve placed my own bets: proximity to the mess. Models are excellent at clean, generic problems, which is exactly why clean, generic tools are dying first. They’re much worse inside a specific company’s reality, where the data is half-migrated, and the billing logic lives in one person’s head. Software wired into a customer’s actual numbers and operations is far harder to replace with a chat window than a PNG converter. That’s the thinking behind Vectig, my cash flow forecasting product, which only earns its keep because it sits on a founder’s real financials. It’s also why my consulting work happens embedded inside clients’ stacks rather than selling seats to something generic.
The part of the thread I respected most was Levels admitting he doesn’t have the full answer. Someone asked what to do next, and he wrote, “If I knew I’d tell you.” I don’t have it either. What I have is a direction: build things that need your customer’s context to be useful, and spend the engineering hours you just got back earning attention you actually own. This newsletter is the second half of that bet, made in public.
P.S. If you run a B2B SaaS and want one AI workflow shipped into your stack in ten business days, that’s what my AI Feature Sprint does. Reply to this email, and I’ll send details.