Trust, Tools, and the Work Only Humans Can Do

Last month I mentioned I'd be sharing more about where the expert economy meets AI. Every month, I'll pull back the curtain on what we're learning at Dewey: the industry shifts we're watching, the tools we're testing, and the real work of building AI that amplifies human expertise instead of replacing it.

With another election day around the corner, this month I've been thinking about what makes people trust information when the stakes are high. Not in theory, but in practice: at 3 AM with a sick kid, at the polling place with questions you're embarrassed to ask out loud, in the quiet moments when you're making decisions that matter.

Trust isn't built in grand declarations. It's built in hundreds of small interactions where you either deliver on a promise—or you don't.

Here's what we've been working on, thinking about, and building this month.

Trust is the Reserve Currency of Our AI Future

Back in 2012, when an advisor told me I was "brilliant," I walked on clouds for a week. These days, I hear "You are absolutely right!" a dozen times a day from Claude—and instead of boosting my confidence, it makes me suspicious.

We're building AI that optimizes for making us feel good rather than being genuinely helpful. And when the stakes are higher than a casual compliment, that's a real problem.

Here's what surprised me: as AI gets more powerful, the people who understand it best trust it less. Wiley's 2025 report shows researchers' concerns about hallucinations jumped from 51% to 64% in just one year, even as usage surged. They're using it more—but trusting it less.

After two years building Dewey, we've landed on three principles for trustworthy AI: "I don't know" is a superpower. Know your boundaries and respect them. Show your sources, always.

These aren't easy choices. In fact, they are crazy, crazy hard. But they are the only way this works long term.

Read the full piece to understand why trust isn't just good practice—it's the foundation of a sustainable system.

You Can't Steer a Ship You're Not On

When ChatGPT launched, I didn't want to build with it. I saw danger everywhere—in every confident wrong answer, in every piece of advice scraped from who-knows-where. I wanted the genie to go back in the bottle.

But this technology isn't going away. I could step back and watch what gets built, or I could help shape it.

For me, that's not really a choice at all. You can't steer a ship you're not on.

This piece walks through why we chose to get on board—not with blind optimism, but with a clear first problem to solve. Read this if you've ever felt torn between the potential and the peril of what we're building.

What Agents Can Do

Ethan Mollick recently gave Claude a complex economics paper and its dataset. One instruction: "replicate the findings." Claude read the paper, sorted files, converted code from STATA to Python, and successfully reproduced the results. Work that would take a researcher many hours, done in minutes.

This is agentic AI in action. Not replacing the expert—freeing them for what they do best.

At Dewey, we see this with every partner. When Dewey handles a reader's election question for Spotlight PA, it frees reporters to analyze campaign promises. When it helps an anxious parent find resources, it lets Dr. Lisa Damour spend more time with patients—time that generates the insights she shares with all of us.

That's not replacing expertise. That's multiplying it.

Read this if you're wondering: What work are you doing that keeps you from your best work?

Spotlight PA's Election Assistant

Building an AI tool for election information is a bad idea, right?

That was my initial reaction when we first connected with Spotlight PA to discuss some big ideas. Election misinformation is next-level bad, of the immoral and dangerous type. But sometimes the scary things push us to be our best

Spotlight PA is doing the hard work of serving Pennsylvania communities with rigorous, unbiased local journalism. Together, we built something worthy of that mission:

Democracy needs trusted information. Journalism needs sustainable tools. Communities deserve both.

Heading into another election season, we're proud to support journalism that takes community service this seriously.

Monologue

I've never been a voice memo person. But Monologue is changing that as I unlock the power of quickly narrating my thoughts.

There's something powerful that happens when I stop pausing to structure my thoughts in text and just let them flow. I'm finding it particularly useful for brainstorming with Claude—speaking shifts my brain into a different mode entirely.

The research backs this up. Talking activates different neural pathways than writing. When you vocalize thoughts, you're engaging motor systems, auditory processing, and spontaneous association in ways that typing simply doesn't trigger. It's why rubber-duck debugging works. It's why thinking out loud helps you solve problems you couldn't crack in your head.

If you've been defaulting to text for every AI interaction, try speaking your next brainstorming session. You might be surprised by what surfaces.

Try Monologue

This newsletter is my way of thinking out loud about where we're headed, but the best insights come from conversations. Hit reply and tell me what's on your mind.

And if you're wrestling with how AI fits into your work or have ideas you’d like to explore, I'd love to talk.

See you next month,
Alex

You're receiving this because you signed up with Dewey Labs - where we build expert-first AI tools. We work with publishers, brands, researchers, and non-profits to create search and Q&A experiences grounded in trust and accuracy. We’ll be sending a monthly dispatch on the future of the expert economy like this one, along with occassional product update.

Upgrade to paid