GREP.AI (YC F26)

Use to create self-improving agents to automate ops work that slows down your business. Grep AI agents get smarter and cheaper with every run.

[GREP.AI (YC F26) @grepdotai]

https://t.co/NoLDjQVwGP

[Fabrizio Serafini @fabserafini]

Love this!!! @grepdotai super smart "right model for the right task"

[AJ Asver @_aj]

Day 2: One of our most common requests from customers was to automate internal tools and cloud software that didn't have APIs or MCPs. Last week we launched that feature in @grepdotai.

Today we're open-sourcing the harness feature that makes it work as an @pidotdev extension.

[GREP.AI (YC F26) @grepdotai]

Our new and improved MCP is live to wrap up our 5 launches in 5 days!

[AJ Asver @_aj]

Day 5 of our week of launches: our new and improved MCP for @GrepdotAI.

Everything you do with Grep agents now works from Claude, ChatGPT, Claude Code, Codex or any MCP client.

→ Build agents from a sentence: Grep designs the prompt, skills and tools, previews its plan, keeps every version and shares it with your team.

→ Run them from any chat: the agent's form opens in the conversation, you see the price before it starts, steer it while it works and collect every file it makes.

→ Stay in control: every run is capped, and your credits, usage and API keys are one question away.

[GREP.AI (YC F26) @grepdotai]

Don't let repetitive decisions slow you down!

[AJ Asver @_aj]

Every day this week, we're launching a new feature @GrepdotAI to help you automate the work that's slowing down your business.

𝗗𝗮𝘆 𝟮: 𝗗𝗲𝗰𝗶𝘀𝗶𝗼𝗻 𝗼𝘂𝘁𝗽𝘂𝘁𝘀 𝗽𝗼𝘄𝗲𝗿𝗲𝗱 𝗯𝘆 𝗝𝗲𝘃

Customers often ask why our agents don't attach confidence levels to their recommendations. My answer has been that LLM confidence levels don't mean much. The model writes a number after the answer, predicting what a confident-sounding score looks like, with a bias toward making humans happy. RLHF pushes models overconfident.

Decision models like Jev by @typesafeai change that. You declare the options up front. Jev returns a probability for each one in a single pass, plus a confidence score derived from how concentrated that distribution is. TypeSafe trained it with RLCD, which rewards calibration instead of rater approval, so decisions at 0.9 should be right about 90% of the time.

In Grep, you can now build agents that use frontier models for the research and diligence, then hand their findings to Jev to make the final decision against your criteria. You set the threshold: above it the agent acts, below it a human reviews.

[GREP.AI (YC F26) @grepdotai]

Skills issue is no longer an issue.

[AJ Asver @_aj]

Hot off the heels of our YC launch last week, we're launching 1 new feature every day this week in @grepdotai to help you automate repetitive work that's slowing down your company's growth.

𝗗𝗮𝘆 𝟭: 𝗖𝘂𝘀𝘁𝗼𝗺 𝗦𝗸𝗶𝗹𝗹𝘀.

You can now create custom skills in Grep for a procedure/process that you use across multiple agents. Grep uses deep research to create the skills prompts and supporting context files and puts into your Skills Library so you can reuse it anywhere.

You can use this to skillify your own institutional knowledge too - just attach the relevant docs, and Grep will create the skill!

[Miguel Ríos Berríos @MiguelriosEN]

Love this. One of the tools I built for our team @grepdotai is an artifact server. It has similar concepts, except it can also publish back to the Claude Code sessions through @herdrdev. Here's a short demo of how we use it.

[AJ Asver @_aj]

We're building self-improving agents to automate tedious, repetitive operational work @GrepdotAI. At IG, a UK online brokerage, Grep agents review compliance alerts, helping over 80% of traders onboard without manual intervention. IG is now a FTSE 100 company.

Operational work often means following a specific procedure, gathering information from different sources, checking it against what's needed, and deciding what to do next.

The steps are repetitive, but no job is the same, making the work tedious to do by hand and hard to automate with workflows alone. Meanwhile, using frontier agents to reason through every task is often too expensive to scale.

https://t.co/EXddQImgCo gives you an agent workforce that:
-> Takes minutes to create and are domain experts with the right skills and tools from day one.
-> Learn from every run, moving repeatable steps into code and workflows and saving frontier intelligence for exceptions and unfamiliar cases.
-> Get smarter and cheaper as they scale using our AgentRun harness.
-> Runs on our managed agent platform with secure sandboxes, audit logs, evals, cost controls, team sharing, and access controls included.

Grep AI agents are available on demand and can run on schedules or in batch. Access them via the console, MCP, API, or SDKs that integrate into your existing workflows.

Our customers, including IG, Airwallex, Worldpay, FV Bank, and Citizens Bank, have saved more than 250,000 hours of operational work with Grep Al agents across onboarding, doc verification, compliance, underwriting, operations, HR, GT ops, market research, and business intelligence.

If operational bottlenecks are slowing your business down, create an agent workforce with Grep to automate them in minutes.

P.S. We're in the @ycombinator Fall 2026 batch!

[Miguel Ríos Berríos @MiguelriosEN]

Friends: help us boost our YC launch in LI. Leave a comment here: https://t.co/epmdGIMmr5

[Miguel Ríos Berríos @MiguelriosEN]

I've been slowly steering Codex to get this done. Finally one device to rule them all. Coding from my Steam Deck on the go and at my desk. Built with Omarchy Quattro (@dhh) and @herdrdev.

[Miguel Ríos Berríos @MiguelriosEN]

Early observation: Opus 5.5 is completely changing the way I work.

1. For large tasks, I'm spending more time with Fable upfront: understanding the problem, reading discovery docs, TDDs, implementation plans, etc. For smaller ones, I just use Opus 5.5 to do the next steps end to end.

2. I give Fable a version of this prompt: "You are the tech lead manager. You have a team of very capable staff and senior engineers. You will give each of them a clear subsystem of this task. They will write code, review it, run e2e tests, iterate, and deliver it to you. You then validate it or send it back with feedback until it's done. When all the work is done, you put the system together and run e2e tests. Go back to the engineers if needed."

3. Fable spawns many Opus 5.5 subagents to do the work.

4. I end up with a stack of PRs to review. We have a "risk" system with rules that either auto-merge after automated checks pass (including autoqa) or require human review. At this point, most PRs auto-merge.

5. The result: almost twice as many PRs merged with no noticeable issues, from the same number of prompts. Since PR size has stayed roughly the same, I read this as getting far more done per prompt. The obvious trade-off is that I'm reading less code. I'm OK with that.

Big props to @chrisrodz35 who put our factory to work so we can do this responsibly.

[GREP.AI (YC F26) @grepdotai]

We've been shipping agents for a while. Here's what we've learned along the way...

[AJ Asver @_aj]

We’ve spent the past three years building agents for real production work. Our architecture changed a lot along the way. Some of those changes worked. Others produced expensive and occasionally painful lessons. Read this to learn from our mistakes. https://t.co/vio067aNWO

[AJ Asver @_aj]

A customer asked me last month how other companies are rolling out AI agents. What's working, what it costs them, and where things stall after the pilot.

I answer some version of that question on calls most weeks. I've never written any of it down to share with others, so I'm restarting my Substack newsletter.

I started the original newsletter, The Hitchhiker's Guide to AI, a few weeks after ChatGPT launched, when everyone was still working out what generative AI was. It let me learn about the space as an outsider looking in. Writing about AI every week eventually led me to the conviction to found an AI startup, and the writing took a back seat. That was three years ago.

The newsletter is now back as The Operator's Guide to AI because I'm no longer an outsider looking in, and neither are you. We're all insiders, trying to build, deploy, and scale AI systems, and I want to share as much as I can about what we're learning along the way @grepdotai.

But there are a million different AI publications, so why should you subscribe to mine?

1/ We've been deploying agents into production since 2023, worked with every generation of models, and made plenty of mistakes along the way. I'll be sharing a lot from first-hand experience, not only of deploying agents at large enterprises, but also of how we use AI internally too.

2/ We speak with large enterprises all the time about their AI transformation efforts: no one has really worked it out, and everyone is a lot earlier than you think. I'll be sharing what we're seeing on the ground.

3/ There's a lot of noise in the AI space, with new models launching every week, new capabilities being announced, and endless fundraising announcements. I want to cut through the noise and focus on what practically matters for folks deploying AI and agents at scale.

If that sounds interesting, you can subscribe below...

The first post on why agents fail in production goes out this week.

[AJ Asver @_aj]

We’ve been busy!

[GREP.AI (YC F26) @grepdotai]

@MiguelriosEN @brieflyin LFG! 💯

[Miguel Ríos Berríos @MiguelriosEN]

In the past few weeks we @grepdotai have launched:

- Make-your-own managed agent
- Loops and monitors
- Media agents
- Our MCP
- A bunch of UX updates and more.

I’m recapping what we’ve shipped in a demo. Showing y’all how I built @brieflyin using our managed agents.

[Miguel Ríos Berríos @MiguelriosEN]

Just added a escalation tool to our @grepdotai cloud agents. I saw a Tweet about this earlier this week and built it. With the right controls in place, this will be super useful.

[Miguel Ríos Berríos @MiguelriosEN]

We released custom agents in @grepdotai this week. I have been using them for long-running tasks. In a couple of hours, Fable used them (through our MCP and @cloudflare's ) to build https://t.co/cnOyDwNsBS. News + podcasts on multiple topics: AGI, San Francisco, US News, etc. Check it out.

[GREP.AI (YC F26) @grepdotai]

Custom agents are live in https://t.co/YJ2Ql1Z0S1! Check them out.

[Miguel Ríos Berríos @MiguelriosEN]

We released custom agents in @grepdotai this week. I have been using them for long-running tasks. In a couple of hours, Fable used them (through our MCP and @cloudflare's ) to build https://t.co/cnOyDwNsBS. News + podcasts on multiple topics: AGI, San Francisco, US News, etc. Check it out.

[AJ Asver @_aj]

At @ParchaHQ we saw coding costs explode so we built our own IDE that does routing at the harness level e.g. Claude Code for planning, Codex for long running execution, Cursor for bugs. Plus custom routing, remote sessions, a browser inspector and voice interface.

Now it's open-source! 👇

[GREP.AI (YC F26) @grepdotai]

Full analysis https://t.co/v5VDdMhptm

[GREP.AI (YC F26) @grepdotai]

Congrats to our neighbors @UseCorgi on a historic raise today. We ran the numbers!