Months before even ChatGPT was released @gabepereyra co-founded @harvey with the bold premise of automating legal work with AI. Today, Harvey is deploying agentic workflows inside some of the world’s most complex and regulated legal environments. I'm very excited to be interviewing Gabe tomorrow as part of the Greylock Change Agents series - link to attend in comments.

sometimes the best ads are free @resolveai

Founding Beacon with @nilamg has been the most intense but also fulfilling time of my life. In just a year, we’ve grown to a team of 30+ and acquired dozens of mission-critical software businesses that quietly power everyday life, helping them bring AI to real-world industries. We are deeply thankful to the entrepreneurs who have entrusted us with their life’s work and our partners at General Catalyst, Lightspeed, D1 Capital, MSD & BDT, and Sator Grove, along with our angels and advisors. This $250M Series B fundraise enables us to increase the scale of our ambitions to build a generational AI holding company. We are hiring across many roles! Check out our careers page.

We're ~warming~ our @SierraPlatform SF HQ next Weds, 11/12. Come celebrate with us and our friends @sequoia! Think tacos, boba, and guest appearances from @claybavor and @btaylor. RSVP at luma.com/u7wx9zst?tk=Bm8EB1

Excellent chat with @gabepereyra from @harvey on last week's Change Agents - full episode to come soon Upcoming @GreylockVC Change Agents includes the CEOs of Glean, Cresta, and Box - DM for invite.

🚨HIRING: Founding Engineers in SF Greylock led the seed in a startup at the intersection of AI x R&D Founders previously built software at SpaceX / Apple / hypergrowth startups and did cutting-edge research in AI and particle physics. They’re tackling a massive market that represents trillions in R&D spend. DM me if you want to help build the foundation of the next industrial revolution.

Why does being the fastest on NVIDIA for gpt-oss matter? 1. Many companies are looking for a US-based open source model, and GPT-OSS is the best out right now 2. Custom hardware providers show up well on benchmarks, but capacity is a real concern for companies at scale 3. The research that goes into making something like gpt-oss faster often ripples into other models, so I wouldn't be surprised to see more huge results from @basetenco on the way 👀

See link to attend here: luma.com/sibnrt5e

🚨 NEW: Change Agents with Resolve AI - AI Agents for Complex Software Engineering Most of the AI tooling conversation right now centers on code generation. Cognition, Cursor, Claude Code....these tools are genuinely useful, but engineers are drowning in production issues. At our latest @GreylockVC Change Agents, @maynkag and @rushins from @resolveai walk through the reality most engineering teams live through: with AI coding tools "you can crank out code much faster, but teams aren't really shipping to production that much faster." 1/ The real problem isn't code generation. Mayank saw this at Splunk where engineers spent only 10-20% of their time writing new features. The rest was fighting incidents, navigating tribal knowledge across hundreds of microservices, and operating tools that don't talk to each other. 2/ Production systems are different from code. Your codebase is static. Production is emergent behavior from infrastructure, deployments, traffic patterns, to tribal knowledge buried in Slack. As Mayank stated, the issue is "a coding agent cannot predict these things without the deep understanding of what's actually happening in production." 3/ The scale required is massive. Resolve is now consuming a trillion tokens per quarter—1% of all Azure usage. That's running Stack Overflow from scratch twice per week. "Vibe debugging" is the missing piece, Rushin's term for getting clarity on production systems the way vibe coding accelerates writing code. Timestamps: (0:30) - Intro (1:00) - Journey to Resolve (4:40) - Rushin’s shift from consumer software (6:04) - Challenges AI can’t solve in production (7:49) - AI in production vs. experiments (9:14) - Production problems (10:50) - Measuring outcomes with agentic AI (12:54) - Coding agents and production risks (14:09) - Vibe debugging (19:20) - Capturing tribal knowledge (20:15) - Codegen tools and production problems (22:31) - Skills needed for agentic AI teams (24:53) - An AI-first organization (26:43) - Audience Q&A

The best marketing stunt since the invention of "Member of Technical Staff" is the rebranding of the Forward Deployed Engineer. Whether you call it Agent PM, Deployment Engineer, Agent Engineer, Member of Technical Staff, or good-old FDE, we are talking about the same thing: an engineer who spends real time with customers. This might be the best role for aspiring founders. You're technical enough to build, but you're in the room when customers explain what's actually broken. No other role allows you to have that type of customer-facing exposure so early in your career. This role has become critical for the fastest moving AI companies, who are putting engineers in front of customers who can configure, customize, and troubleshoot in real-time. The playbook works. Greylock portfolio companies @basetenco, @cogent_security, and @tryramp are going to talk about how they think about these forward-deployed roles and how it has been critical to their growth at our event on October 14th. DM for Invite

What do Baseten, Cogent Security, and Ramp have in common? 1/ Engineers love working there 2/ They are all @GreylockVC portfolio companies 3/ and.....they're all actively hiring FDEs You can come learn more about how these iconic companies think about FDE roles at our event on October 14th. DM for invite. @basetenco @cogent_security @tryramp

Many thanks to @CorinneMRiley and @GreylockVC for having me on Change Agents. Could not have asked for a better panel alongside @vineete_5, discussing how AI is ushering in new paradigms in cyber today! Check out the full conversation: piped.video/watch?v=WyHE6s7J…

New Greylock Change Agents with @Abnormal_AI_Inc and @cogent_security We spoke with @evanreiser (CEO of Abnormal AI) and @vineete_5 (CEO of Cogent Security) - two of the leading AI-native security companies adopting AI Agents to fight cyberattacks. Cybersecurity is rapidly evolving as AI-driven threats grow more sophisticated and defense strategies shift from static rules to intelligent systems that can think, reason, and act. The role of Agentic AI in cyber cannot be understated - but how is that taking shape? (0:00) - Intro (1:02) - About Abnormal and Cogent (2:36) - How Even and Vineet met (4:14) - Attack landscape shift (6:25) - Agentic reasoning in products (10:23) - Example use cases (13:01) - Adopting agentic security (14:44) - Enterprise adoption (17:15) - Trust and humans in the loop (21:28) - Future of agentic capabilities (23:32) - 5-year outlook (25:47) - AI impact on workflow (29:42) - Team building and culture (33:12) - Leadership traits (35:01) - Learnings from Abnormal cc @GreylockVC

On the future of evals: @ankrgyl has been working on evals since 2016. The biggest difference since then is that the amount of effort it takes to incorporate what you learn from the eval into the next iteration is very little. “We went from extremely slow updates that were very manual, to very fast updates that are very manual, to even faster updates that are partially automated. The way we as humans interact with evals is going to shift from looking at a dashboard, walking away with some inference, and then typing something to test. It’s going to be more like an LLM system suggesting and contextualizing what we should change and why.” cc @braintrustdata

Is evaluating agents more complex than evaluating non-agentic AI software? @ankrgyl says: NO. “Agents are a very natural evolution of AI software. The agentic system that our most sophisticated customers build is actually dramatically simpler than previous generation of software. Yes it’s more powerful, the model may be doing more steps independently, the user experience may end up being more sophisticated, but the actual code and the logic in the code is dramatically simpler." cc @braintrustdata

"When I run an eval I look at two things: 1// What are the specific tests that got worse vs my previous eval - are these things actually worse, and if so let me play with them. Or I look at say 'what the hell' and change the scoring function so I make sure my human intuition is encoded into a strictly better system than before. 2// Improvements: you should look at with the exact same skeptical eye with which you look at regressions. Did it actually improve or is it a fake improvement?. Often, when you’re getting started most of the improvements are fake” cc @ankrgyl @braintrustdata

You can see the full recording here: piped.video/watch?v=J0NdPARq…

NEW Greylock Change Agents: Evaluating Agents with Braintrust There's been a lot of discourse around evals recently, so it's timely to drop the recording of Change Agents with @ankrgyl on the topic of how he thinks about Evaluating Agents. Timestamps: (0:00) - Intro (1:22) - Building Braintrust (5:19) - Evaluating agentic systems (11:04) - Experimentation with evals (13:18) - Adopting evals (17:09) - Evals and complex agents (21:19) - Custom scoring (25:02) - LLM as a judge (26:44) - Human-manual evaluation (30:07) - How the best in-class teams manage evals (33:57) - Future of evals for agents (37:38 ) - Audience Q&A

Congratulations @jaltma and the whole Alt Cap team, who are truly some of the best in the biz 🐐

“The key is having good intuition, being willing to go out on a limb, building fast, learning fast, and killing things when you need to.” Following our Series D raise, our Co-founder and CTO @amiruci walks through why he bet early on inference, how we’re scaling through generative model hypergrowth, and his advice for fellow founders.