Vibe Check: I Canceled Two AI Max Plans for Factory’s Coding Agent Droid

Get up to 60 million free tokens to spend on any model—GPT-5, Sonnet 4.5, Opus, etc.—by attending our Droid Camp for paid subscribers tomorrow, October 31, at 12 p.m. ET. Factory AI’s head of developer relations Ben Tossell and Every’s Dan Shipper, Danny Aziz, and Kieran Klaassen will walk you through how they use the tool. RSVP if you’re a paid subscriber, or upgrade your subscription to attend.—Kate Lee

Was this newsletter forwarded to you? Sign up to get it in your inbox.

Picture reorganizing your filing system while people are still thumbing through the folders. That’s what it’s like to move or restructure data in your production database—the live system storing all the user data that powers your product.

Now imagine that you’re trying to reorganize the system that runs your entire product while you’re balancing your laptop on your lap in the back of a moving car. That’s what it felt like when a production database migration I was working on for Spiral, the AI writing tool I’m building at Every, started failing on repeat while I was in the back of a car headed to IKEA. Not where you want to be debugging, but when things break, they don’t wait for you to be at your desk.

I’d started this migration, which involved restructuring how conversations, drafts, and messages are stored inside Spiral’s database, with Claude Code. The problem was that it kept circling the same dead ends of edge cases in production data that Claude Code couldn’t anticipate. So I opened Droid, set to the same Anthropic model, Opus 4.1, and asked for a fresh migration. It wrote one, applied it in a single shot, and explained each step as it went.

Phone calls suck.

Especially when you’re dialing a big company. You call with a simple question… and what do you get? Hours on hold. A clunky phone tree that leads nowhere. Or a robotic voice that couldn’t sound less human. And after all that? You still don’t get the help you needed.

That’s why Bland, the world’s first self-hosted voice provider, is building their AI agents. Now, when you call a company powered by Bland, you get:

1. A voice that sounds authentic, human, and engaged
2. Instant responses (<500ms)
3. Actual helpful answers (imagine that)

It’s no wonder more and more enterprises are rolling it out—especially once they find out it’s the only provider where their data is actually secure. So… expect to be talking with Bland’s agents a lot more soon.

👉Feel free to try it out yourself: www.bland.ai/talk

Or if you’re an enterprise, schedule a custom demo: https://www.bland.ai/dem

Want to sponsor Every? Click here.

The thing about coding with AI in 2025 is that no single model is best at everything. Sonnet excels at complex code cleanup (i.e., refactors). GPT-5 is fast at prototyping. Haiku is great for quick fixes. But switching between tools to access different models involves losing your context, relearning commands, and breaking flow every time you need a different brain.

Droid lets you switch between models made by different labs without switching tools—and somehow makes each model work better in the process. It’s the first agent wrapper—the software layer that packages an AI model into a usable tool—I’ve tried that feels like it was built for how senior engineers work: running parallel sessions, composing workflows across models, and staying in flow when the work demands it.

It’s been a few months since I started tinkering with Droid, and I’ve canceled my Max plans for both Claude and ChatGPT. Droid has replaced them both. Let me tell you why...

1. The first command line agent that runs both Anthropic and OpenAI models
2. How Danny burned through 200 million tokens in a week and ditched his existing subscriptions
3. “The strongest planning mode I’ve used”