What’s 🔥 in Enterprise IT/VC #488

When Product Velocity Breaks the Company

Agents unlocked engineering velocity we’ve never seen before. But here’s the problem nobody is talking about: the org hasn’t caught up.

Engineering used to be the bottleneck. Now it’s not. The constraint has shifted to everything around engineering - security review, launch decisions, sales enablement, customer comms. GTM can’t keep up with weekly releases. Sales is still absorbing last week’s launch while engineering ships three more. Your champion at a customer account has no idea what you’ve shipped since their last QBR.

This isn’t instability. It’s acceleration without a shock absorber.

Think about it as three derivatives. First derivative: agents write code faster than any human team ever could. Second derivative: more AI-generated code means more code to review and more code to secure — and none of it is perfect. You don’t want the same model that wrote the insecure code also signing off that it’s secure. We already have a live example of what happens when agents push code to prod without the right controls in the loop.

But the third derivative — and the one nobody is talking about — is the organizational problem. Velocity is no longer a code problem. It’s an org design problem.

The fix isn’t slowing down. It’s building the organizational infrastructure to match the speed. A few things the best teams are starting to figure out:

Separate the ship calendar from the launch calendar. Engineering ships continuously. GTM picks 2-3 moments per quarter to go big. Everything else flows quietly underneath.

Replace update meetings with live demo syncs. Fifteen minutes, live product, every week. Sales instantly knows what to talk about. Marketing sees the story. Kills the “I didn’t know we launched that” problem dead.

Build an AI-queryable customer portal. Your champion shouldn’t have to wait for a QBR to know what changed. “What shipped since my last login that affects my use case?” should have an instant answer.

Hire agent red-pilled across every function. Not just engineering. Finance. Marketing. Sales. Every new hire needs to understand agency - how to use it, how to build with it, how to manage teams around it. For leaders, the question is simple: how are you going to build teams differently? Smaller, fully bought in, agent-native from day one.

The startups that nail this turn engineering velocity into customer expansion. The ones that don’t turn it into customer confusion.

As always, 🙏🏼 for reading and please share with your friends and colleagues!

---

Scaling Startups

#we still need physical labor…it will take awhile for the robots to come but striking difference with white collar jobs which are still higher paying - the world still needs plumbers

# I don’t know about you, but the more agents I deploy the more context switching i have to do during the day and the more my head hurts - yeah, lots of output but wow (👇🏻 for full article)

#🤔 yes and no - depends how you describe pure software but once again goes back to what I’ve repeatedly said - well worth reading the comments

#always great to learn how the fastest growing AI native companies scale and still maintain speed - all about culture and then how it manifests itself into action, here’s Lovable’s founder talking about what’s ahead as it scaled from 20-150 people - love how many ex-founders it has hired…similar to Ramp model

---

Enterprise Tech

#great post if you’re a Zirpicorn 🦄 and need to reinvent yourself - how Intercom did it and now at $400M ARR 🤯 - thank you for sharing Eoghan

#this is really good and 💯

#are harnesses the new moat? lots of discussion on harnesses - this post argues that AI agent success hinges on the “harness”—a simple execution loop with primitive tools, error handling, and controlled information flow—rather than the underlying model, as evidenced by Claude Opus 4.5’s performance jumping from 42% to 78% on CORE-Bench solely via harness tweaks.

but wait a second…turns out the more you build into a harness the less flexibility you have as every new model incorporates some of your proprietary harness🤔 from How to be a world class engineer from Systematic Long Short

#some days my head feels like it will explode having to context switch to keep my agents running and while i can run many of these through Claude, i also don’t want to - we will have agents from every single SaaS vendor like Slack and Salesforce and ServiceNow and we will create lots of our own. I was thinking more for personal use - having my own orchestration layer to understand where they are, what systems they are connected to and what my agents are doing. Apple could have owned this right in OS but so far have fumbled. Click 👇🏻 for folks sharing so many interesting ideas and ways. Of course building an enterprise version is a whole lift and will be complicated in terms of who wins.

#great post from the founder of Cursor on the Third Era of AI Software Development and limitations of local machines versus cloud based VMs

#and uh, well, rumors of Cursor’s death seem to be just that 🤯

#memory is not a storage layer, needs to be more intelligent, read how CrewAI built its system (a boldstart portfolio co)

#lots of great discussion on whether or not thin wrappers can become thicker ones and who is long for this world and not

#Decagon valued at $4.5B saying I’m not just a pretty thin wrapper for customer support and that it’s trainings its own models and not just a bunch of FDEs like Bret Taylor’s Sierra

What businesses actually need is a natural language interface and an execution engine capable of capturing arbitrarily complex logic across wildly different customers. That requires training custom models to meet real performance requirements. It also requires building the surrounding platform tooling — versioning, analytics, simulation environments, automated self-improvement — so that less-technical teams can build and optimize their agent without engineering support.

To accomplish this, we’ve assembled high-agency teams of cracked engineers, applied AI researchers, and former/future founders.

#independent guardrails still needed to secure agent generated code

#if you’re a layer between the LLMs and the data, better build some pretty insane embedded workflows

---

Markets

#sad but true commentary you should read on why more layoffs to come…

#lessons learned from a legendary macro investor Stan Druckenmiller - so many 💎

“Contrarianism is overrated. The crowd’s right 80% of the time.”

The trick is being bleeding F8#*ing early, before the crowd and hoping a few Inception bets become the next 80% winners.

ties in nicely from this famous Stan quote, not in interview, but one to 🤔 about:

"If you’re going to be a great investor, you’re going to have to be willing to have people think you’re stupid."

I often feel that way with many of inception investments. Any how, watch and absorb, and here’s a great post if you want 10 of his key thoughts from the interview.