Developer Built Her Own LinkedIn AI Tool in Two Days

There's a particular kind of frustration that hits when you're a developer and every tool you try sucks. You know exactly what you want. You can visualize the workflow. But everything on the market is either too generic, too robotic, or solving the wrong problem entirely.

That's where Colleen Schnettler found herself with AI-powered LinkedIn tools. She's a Ruby on Rails developer who taught herself to code nights and weekends while working what she calls "a decidedly mediocre job in a different field." She's worked at ClickFunnels, runs a SaaS called Simple File Upload, and is now launching a technical marketing consulting business. And every single paid AI LinkedIn tool she tried? Hated them all.

So she did what developers do when the market fails them: she built her own in two days.

The actual workflow is embarrassingly simple

Here's how Schnettler's system works: She uses Super Whisper (though she's "been going back and forth" between that and Whisper Flow) to capture voice notes throughout the day. Her custom app watches that folder, reads the transcripts, and auto-generates LinkedIn posts.

That's it. Voice note to LinkedIn draft. No clicking through templates. No fighting with prompts in some bloated interface. Just thoughts → text → post.

The real example she walked through: A founder friend sent her a Slack message about Lenny's podcast with Molly Graham. While responding, she just... voice noted her reaction. "I'm obsessed with this new podcast. In the first 8 minutes, Molly Graham says things like, 'I only do jobs I'm highly unqualified for.'" Captured her thoughts about why that resonated, why it matters for how she works.

The system automatically turned that into a LinkedIn post. "And it might be terrible, right?" Schnettler admits. "But it's a first step, right? It's a way to capture ideas for LinkedIn."

That's the actual value proposition here—not that AI writes perfect posts, but that it removes the friction between having an idea and doing something with it. Most LinkedIn post ideas die not because they're bad, but because you're "usually just too lazy to actually go and post it," as she puts it.

How she actually builds with AI code generators

Schnettler uses the Compound Engineering framework from Every—a set of slash commands for Claude Code that creates structured plans, executes them, and reviews results through specialized sub-agents. It's free, installable from the marketplace, and she landed on it because "it feels the simplest."

Recently, she added a full chat interface to her LinkedIn tool in one session. Split-screen layout: chat on the left, editable post on the right. That's a legitimately complex feature—real-time chat interaction that needs to modify content in a separate pane.

Her process: She typed one prompt. "Add a split screen chat interface to the LinkedIn post show page where users can discuss and refine their post with AI assistance. The chat window appears on the left with the editable post on the right. So this is this was my prompt I think almost verbatim."

Claude generated a plan document. Not code—a plan. What needs to happen, key design decisions, implementation details. She reads it, catches the things it got wrong ("it got a few things wrong the first time"), has it revise, then executes.

"I just kind of look at it high level," she says about reviewing plans. For small stuff—changing a button color—she skips the plan entirely. But anything with real scope gets the full treatment.

The part that actually matters: she doesn't trust any of it

Here's where Schnettler's workflow gets interesting. She runs multiple Claude Code instances in parallel. Two max—"I hear the stories of people who are doing 10 and I'm like I don't know. I don't know how you can keep that context."

When Claude gets stuck, she bounces to GPT-5 in Windsurf (her IDE of choice). Not because one is definitively better, but because sometimes you need a different perspective on the same problem. She showed an example: Claude was scanning all files in her Super Whisper folder, which is fine at 100 files but "when I have 10,000 files, this is going to be a problem." Claude missed that nuance. GPT-5 caught it.

And here's the critical part: "I'm actually going to take time to be like, is this a real problem?" When the review agent flags issues as "critical," she looks at them with her own eyeballs. She almost never tells AI to "go fix them all." She goes through issues one by one.

Why? Because "there is still a feel to an app that AI can easily miss," as the video notes. Schnettler barely opens her IDE anymore ("Not unless something breaks or something feels wrong"), but she absolutely still clicks through every feature in the browser before merging. That developer intuition—"something feels wrong"—can't be automated away.

What's actually happening here

Schnettler's workflow reveals something about the current state of AI coding tools: they're incredibly powerful for execution, genuinely mediocre at judgment.

The Compound Engineering framework runs 11 sub-agents—Security Sentinel, Architect, Strategist—and they'll churn through implementation details faster than any human. But they'll also miss obvious scaling issues. They'll generate technically correct code that feels wrong. They'll create features that work but don't fit.

The leverage isn't in replacing developer judgment. It's in removing the grunt work that sits between "I want this feature" and "here's a working prototype I can actually evaluate."

Schnettler's LinkedIn tool exists because she could build it in two days instead of two weeks. Not because AI wrote perfect code unsupervised, but because it handled enough of the mechanical work that she could focus on the parts that actually require taste: Does this capture ideas the way I actually think? Does the chat interface feel right? Should posts be tagged by framework as they're generated?

"I'm just making this up as I go," she laughs at one point, feeding her personal LinkedIn strategy from Notion into Claude. "Lead the way, Claude. Let's go."

That's the real workflow. Not AI replacing developers. Developers using AI to make rapid iteration cheap enough that they can just... try stuff and see what sticks.

—Tyler Nakamura, Consumer Tech & Gadgets Correspondent