Skyvern (YC S23)

My AI writes a first draft. Then I send other AIs to tear it apart. A single agent confidently producing one draft is how you get slop. An agent forced to defend its work against other agents is how you get something worth reading.

New rule: if it isn't recorded, it didn't happen. Harsh, but it's exactly how your AI sees your company.

I post 5x a week and write none of them from scratch. An agent reads my last 20 customer calls every morning and drafts them. I just edit and publish. Consistency was never a discipline problem. It was a context problem.

The best "prompt engineering" I did this year: gave an agent read access to our database. Stop wordsmithing the question to squeeze a smarter answer out of thin context. Give it access to the answer.

Everyone sells AI on speed. Speed is the boring part. The real unlock is memory: an agent holds every call, ticket, and Slack thread at once and surfaces the right one at the right moment. I forget my own week by Friday. It doesn't.

I record every call I'm on. Internal, external, even 1:1s with my cofounder. I'm genuinely not sure it's okay. But an agent pulled a customer's exact words months later and handed them to the engineer fixing the bug. Where do you draw the line?

Remote companies have an unfair AI advantage: Every meeting is recorded In an office, your best context is spoken and gone in seconds Your agents can only act on what got written down.

Your AI doesn't produce slop because the model is dumb. It produces slop because you handed it a task with none of the context a human would need to do it well.

We banned DMs at our company. Every question goes in a public channel instead. A DM is a private context leak: knowledge two people can see and your agents never will.

My team kept calling my work slop. It was an AI agent writing our product specs. Complete slop The fix? more context

An AI wrote one of my LinkedIn posts last week. I didn't write a word of it. It got me a sales lead

Every AI agent startup is selling "we're 10x faster." Every production buyer I talk to is saying "I don't need faster, I need it to actually work."

We paused all feature work this week to focus on stability. Customer support tickets dropped 35% in 4 days. The roadmap can wait. Bleeding can't.

Healthcare AI founders: how long does your BAA take to negotiate? If it takes > 9 days you're in trouble

Everyone obsesses over model quality. We spent 6 weeks rewriting our 2FA handler. That's why customers stop churning. The reason your AI agent dies in prod isn't the model. It's that you don't have a credential vault.

blocked 30 min/day this month to manually ping every active customer. 1 in 3 had something quietly broken nobody had flagged. slack connect isn't a relationship. a CSM dashboard isn't a relationship. someone has to actually type the words "hey, how's it going."

genuine question for AI agent founders: can your customer take the code your product generated and run it themselves if they churn? if no, they're going to ask. and your answer is going to decide your enterprise win rate in 2026.

unsexy take: the most valuable AI agents in 2026 won't automate code. they'll automate state nurse credentialing, e-visa filings, and tax registration in 50 states. the boring verticals nobody demos at conferences are eating the world quietly. i'm watching it happen on every call.

Asked Claude to generate memes for future posts First try? Hilarious

No one wants their browser agents to book flights for them What they actually want is to automate mortgage portals or do medicare verification