🎧 Prefer to listen?

Miles Wang just left OpenAI to build a $2 billion startup. His strategy? Not inventing new drugs — finding new uses for drugs that already exist. Compounds that passed safety testing but failed for their original purpose. The hard part (proving they’re safe) is already done. He just has to figure out what else they can treat.

And it hit me: this is exactly the move most solo builders are missing.

We’re all sitting on tools we’ve already paid for, already learned, already integrated into our workflows. But instead of squeezing more value out of them, we keep refreshing Twitter for announcements about the next model drop. I wrote about Wang’s original move when the news broke, but the real lesson isn’t about biotech. It’s about how you think about the tools you already have.

The “new model” trap

Every few weeks, another AI model ships. GPT 5.6, Claude Opus something, Gemini Ultra Whatever. And every time, the same cycle plays out: hype, benchmark comparisons, “is this the one that changes everything?” think pieces, and then… you use it for the same three things you used the last model for.

Sound familiar?

I’ve been there. When Claude Cowork launched, I spent a weekend setting it up. When Gemini got image generation, I tested it for hours. When OpenAI announced Codex hardware, I wrote about what it meant. But here’s the uncomfortable truth: most of my actual productivity gains didn’t come from new tools. They came from finding new ways to use tools I already had.

Wang’s startup isn’t built on a new kind of AI. It’s built on applying existing transformer architectures — the same tech behind ChatGPT — to biological data. The model isn’t novel. The application is.

What drug repurposing looks like for your business

In pharma, drug repurposing works because the compound has already cleared the hardest hurdle: safety testing. The expensive, time-consuming part is done. You just need to find the right match between an existing molecule and a new disease.

For solo builders, the equivalent is this: your tools have already cleared the hardest hurdles too. You’ve already learned how to prompt them. You’ve already connected them to your workflow. You’ve already paid the subscription. The expensive part — the learning curve, the integration, the trial and error — is behind you.

Now find the new applications.

Here’s what that looks like in practice:

You’re using ChatGPT for writing. Have you tried it for data analysis? Upload a CSV, ask it to find patterns, generate charts. I covered this when I tested 10 different AI writing tools — but writing is maybe 30% of what these tools can actually do.

You’re using Claude for long documents. Have you pointed it at your competitors’ websites and asked for positioning analysis? Have you fed it your customer support emails and asked for pattern detection? That’s not a new tool — it’s a new use for an existing one.

You’re using Make or Zapier for basic automations. Have you built multi-step workflows that chain AI calls together? My automation pipeline started as a simple email-to-Notion flow. It now handles content scheduling, client follow-ups, and research aggregation — all with tools I was already paying for.

The compounding problem nobody mentions

Here’s why repurposing beats upgrading: compounding.

When you switch to a new tool, you start at zero. New interface, new quirks, new failure modes. You spend weeks getting comfortable. Then another tool drops, and you switch again. You’re always climbing the first hill.

But when you go deeper with a tool you already know, each new application builds on everything you’ve learned. Your prompts get better. Your workflows get tighter. You discover edge cases that save you hours. That knowledge compounds in a way that tool-hopping never will.

The AI tools with the highest satisfaction rates aren’t always the newest ones. They’re the ones people have been using long enough to get good at them.

A concrete repurposing exercise

Try this today. Pick the AI tool you use most. Open your last 20 conversations with it. Look for:

  1. Repeated tasks you could templatize. If you’re writing the same type of prompt every week, save it as a template. Most tools support this natively now.

  2. Adjacent tasks you haven’t tried. If you use it for writing, try research. If you use it for coding, try data analysis. If you use it for brainstorming, try structured decision-making.

  3. Connections you haven’t made. Can this tool talk to another tool you already use? Webhooks are the bridge most people ignore.

  4. Outputs you’re not capturing. Are your AI conversations generating insights you’re not saving? Set up automatic logging. The raw material for your next project might already be in your chat history.

Wang didn’t invent a new molecule. He looked at existing molecules and asked, “What else could this do?” You have the same opportunity sitting in your tool stack right now.

Stop waiting for the perfect model

There’s a specific kind of procrastination that masquerades as strategy: “I’ll build that workflow once GPT 5.6 ships.” “I’ll automate that process when Claude gets better at structured output.” “I’ll set up that system when the tools mature.”

You won’t. Because there will always be another model three months away. And when it ships, you’ll wait for the next one.

The AI price war is making tools cheaper every month. The 7 AI tools I’d learn first if I started over today are all tools that exist right now. The capability gap between “what you wish you could do” and “what your current tools can actually do” is almost certainly smaller than you think.

Wang’s $2 billion bet isn’t that AI will get better. It’s that existing AI, applied to the right problem, is already good enough. He’s not waiting for the next model. He’s repurposing the current one.

The real takeaway

The next time you catch yourself refreshing TechCrunch for AI news, stop. Open the tool you’re already paying for. Ask it to do something you haven’t tried before. Apply it to a problem you’ve been saving for “when the technology is ready.”

The technology is ready. It’s been ready. You just haven’t pushed it far enough yet.

Drug repurposing works because the hard part — safety testing — is already done. Your AI tools have already passed safety testing too. You know they work. You know they’re safe. The only question left is: what else can they do for you?

Start there. The next model can wait.

This post references reporting from TechCrunch on Miles Wang’s startup plans. For the original industry breakdown, see our earlier coverage. For more on getting the most from AI tools, check out the AI stack I’d use starting from zero and stop doing things manually.