Your Company Gave Everyone Claude Code. Now What?

Key Takeaways

• "AI-forward" is mostly an illusion. Companies handed everyone Claude Code and Codex, then routed everything employees built into week-long DevOps queues — "holding employees back without realizing they're doing so."
• The real cost is invisible. The discouraged builders you can count. "The people that aren't encouraged" — the apps nobody ever starts — you can't.
• Most software is about to go internal. At Doublepoint, the gesture-detection startup acquired by Oura, there was no customer-facing product at all: the internal platform was the product.
• The best harness is my harness. Railcode has no drag-and-drop builder on purpose. Build with your own tools — Claude Code, Codex, Cursor, opencode — and deploy in about five seconds to a space your company can actually govern.
• Pull the thread. Yakko found Railcode by re-reading every customer note from his invalidated ideas and matching them against a decade of infrastructure work. The idea was already in his notes.

Listen: Apple Podcasts | Spotify

Introduction

Two days before this conversation was recorded, Yakko Majuri launched Railcode. Most founders would still be refreshing their analytics. Yakko sat down across from Julian and opened with a diagnosis instead: the companies proudest of being "AI-forward" are the ones quietly blocking their own people.

His test for whether Railcode solves a real problem is a single question, and it doubles as a test for every company that rolled out AI coding tools this year:

"Are you unable to keep up with your team's development with AI today? … Are you holding your team back?"

Yakko has spent his career underneath software. He was one of PostHog's earliest engineers, owning infrastructure and internal tooling. He then led infrastructure at Doublepoint, the Finnish gesture-detection startup acquired by Oura. Now he's solo, a member of the Solo Founders Program, and building the thing his whole career has been circling: a secure home for the internal apps and AI agents that employees are suddenly able to build — and mostly unable to ship.

The AI-Forward Illusion

Here's the pattern Yakko kept hearing in customer conversations. A company gives everyone access to Claude Code or Codex. People light up. They build tools that genuinely help them do their jobs — working prototypes, running locally, today. Then they try to share what they built, and the machinery of the company stops them: DevOps review, security questions, no sanctioned place to deploy. Most of his customers had waited a week or more to get something live. Plenty of people never bother.

"There are companies out there talking about being AI-native, and they are holding employees back without realizing they're doing so."

The visible cost is the delay. The real cost is invisible: "There are people that are discouraged, and then there are the people that aren't encouraged. And it's really hard to measure the second one." When nobody around you is shipping internal tools, you don't conclude the process is broken — you conclude that building isn't something you're allowed to do. As Julian put it in the episode, the painful part isn't the week-long queue. It's the thing that never gets built at all.

Yakko's summary of the corporate mood is almost gentle: "We've given everybody Claude Code. We've given everybody Codex. We're AI-forward." But — what's the next step?

Most Software Is About to Go Internal

The bigger claim underneath Railcode is a market thesis: "I think most of the software going forward is going to be internal software."

Yakko lived an extreme version of this at Doublepoint. The company sold gesture-detection algorithms to device manufacturers — there was no customer-facing product to polish. Everything that mattered happened inside: the data platform, the training pipelines, the tools leadership used to steer. Once he built a real home for that internal work, he watched the whole company compound — faster model training, more data collected, better decisions. Then Oura acquired the company. "For some companies," he says, "the internal tooling is the product."

That inversion is easy to miss if you think of internal tools as the boring cousin of "real" product work. Nearly all of the AI codegen wave has gone toward customer-facing features. The internal layer — where the company actually runs — is still waiting for its turn.

The CEO, the CTO, and the Ten-Hour Week

The best evidence in the episode is a story about one small company. The CEO is fluent with Claude Code and wants to build. The CTO trusts neither the security practices of a vibe-coded app nor the idea of handing the CEO database keys. So every internal tool becomes a negotiation: the CEO asks, the CTO builds, a two-hour favor becomes a four-hour project. The week Yakko closed the deal, the CTO admitted he'd spent ten hours just supporting the CEO's requests.

Railcode ended the standoff — not by making the CTO looser, but by giving him something he could actually audit: permissions, authentication, and governed connectors in one place. The CEO now builds everything himself. And the twist Yakko clearly enjoys most: the CTO started building there too, because it was faster than wiring up deployment the proper way.

This is the part of the permissions story that surprised Julian, and probably surprises most people: it's not a big-company concern. The moment a small team has customer data, "who can access what" stops being optional.

The Best Harness Is My Harness

Railcode has no drag-and-drop builder, and that's a decision, not a gap. Yakko watched every product bolt an AI chat box onto its interface and drew the opposite conclusion:

"The best harness is my harness."

People already have their tools — Claude Code, Codex, Cursor, opencode — loaded with their skills, their workflows, their testing setups. Railcode's bet is to leave all of that alone. Build locally with whatever you use. The code stays yours: put it on GitHub, run CI/CD. Railcode is the secure place it deploys to, with auth handled, permissions granular, and connectors — Postgres, Google Calendar, Granola — governed by the company instead of scattered across whoever's Vercel account happened to be logged in.

He demonstrates it live in the episode: notices a demo Kanban app has no dark mode, fires off Claude Code to add it, types railcode deploy, and it's live about five seconds later. Apps built there are immediately usable by agents — build a custom CRM and an agent can operate on it from Slack with no extra work. It's a preview of a flip he predicts with a line worth remembering:

"Right now it's a lot of UIs that have agents, and I think we're going toward a place where it's maybe agents that have UIs."

And where teams today share a single password over Slack for their vibe-coded tools — everyone acting as the same user — "Railcode by default is multiplayer. It already knows who the users are."

The Through-Line

The founding story matters here, because Railcode wasn't a lightning-strike idea. When Yakko joined the Solo Founders Program, he was between ideas — he'd pivoted off one product and invalidated another. What he did next is a repeatable method: he went back through every note from every customer conversation across all his previous ideas, looking for the problem that kept surfacing. Then he checked it against his own history — PostHog infrastructure, Doublepoint's internal platform — and found the fit. Some of the code from his previous product lives on inside Railcode. "In some ways it's actually an evolution."

Julian named the pattern: founders rarely land on the idea by leaping somewhere new. They pull the thread.

The PostHog chapter supplied the operating style, too. Yakko joined because of the company's famous public handbook — "I think a lot of people did" — and a three-day work trial proved the culture was real. No teams larger than six. Products born at hackathons, including the earliest version of session replay, which he was in the room for. What he kept from all of it: iterate fast, and learn from customers actively.

He was still iterating on his launch-week website — the one with a playable roller coaster in it — when he stopped a customer call to ask a question most founders wouldn't: is Railcode fun? The answer, which doubles as Yakko's whole philosophy of work: building things is fun, when someone else handles the parts you dread. "It's fun — and fun is useful."

The Case for Going Solo

It's worth noticing what this episode looks like from the outside: a founder, alone, days after launch, giving the first live demo of his product on camera — no team to coordinate, no message to negotiate, nothing between the insight and the audience. The operating style Yakko absorbed at PostHog — small autonomous teams, fast iteration, learning from customers actively — turns out to work at a team size of one. He was mid-iteration on his own launch website when he stopped a customer call to ask whether the product was fun. That's not a luxury of solo founding; per Yakko, it's the mechanism.

His advice for solo founders is the same idea, turned inward:

"Get in touch with yourself and the things you want to build. That's the way you're going to build something that means something to you. You're going to work harder, stay at it longer, enjoy it more — and consequently, you're going to serve your customers better."

He's a member of the Solo Founders Program, building alongside other founders who chose the same structure — solo together rather than solo alone.

About Yakko Majuri

Yakko Majuri is the founder of Railcode (railcode.dev), a secure platform for building and deploying internal apps and AI agents. A backend and infrastructure engineer by identity, he was one of PostHog's earliest engineers — owning core infrastructure and internal tooling, and joining via the company's famous public handbook — before leading infrastructure at Doublepoint, the Finnish gesture-detection startup acquired by Oura in 2026. He's a member of the Solo Founders Program, and launched Railcode two days before this episode was recorded. Find him on X at @yakkomajuri.

Subscribe to the Solo Founders Podcast on YouTube, Apple Podcasts, or Spotify.

Join our newsletter for weekly insights from solo founders: solofounders.beehiiv.com

Building solo? Apply to the Solo Founders Program: solofounders.com/program