←
Andy Berman7,775 posts
Andy Berman avatar

Andy Berman

@berman66

CEO of @runlayer. Prev: Director of AI @zapier, Founder @getnanit

◷ Joined March 2008▧ 742 Photos and videos
Andy Berman avatar
Andy Berman@berman66·16h

Almost everyone, including modern security teams, feels underequipped. Attackers now use AI agents to find weak spots and exploit them at machine speed. Even if the labs keep bleeding edge cyber features behind gated access, the open-weight models are just weeks behind. At the same time, most companies have handed their own AI tools access to email, code and customer data, and they can't see what those tools actually do with it. If you want to protect yourself: 1) Put passkeys or a hardware key on your email, bank and password manager. Whoever gets into your email can usually get into everything else. 2) Use a password manager and never reuse a password. 3) Go through everything you've connected to Gmail, Drive, GitHub or Slack: AI assistants, browser extensions and "Sign in with Google" apps. Revoke anything you don't use. Each one is a way in. 4) Turn on automatic updates everywhere. If you run a team: 1) List every AI agent and MCP server your people use. 2) Give agents only the access they need, and require a human to approve anything destructive, like deleting data, moving money or pushing to production. 3) Log every action an agent takes. When something goes wrong, you should be able to find out what the agent did in minutes. Today that often takes weeks. If you want to see what attacks on AI agents look like in practice, my DMs are open. x.com/danshipper/status/2122…

the current spate of cyber attacks seems a sign of things to come. if something doesn’t change, we should expect coming attacks to do real damage. i personally feel under equipped to know what precautions to take or how I can help. Who’s doing good work on this?

Andy Berman avatar
Andy Berman@berman66·18h

The best AI platform team I talked to this month has one rule: don't build anything core. They go cloud vendor first, then open source, then a thin layer to stitch them together. I'm not surprised. If it's core, you need to be careful. Everyone goes down this build path because it seemed easy, and then they run into the same problems. Maintain seems to be the hardest part.

Andy Berman avatar
Andy Berman@berman66·Nov 16

Ask a non-technical employee to "build an agent" and they have no idea what to do. Ask them to onboard a new teammate and tell it how to do the job, and they get it. It really is just as simple as telling AI what you want to build.

Andy Berman avatar
Andy Berman@berman66·Nov 16

Personal agents vs Platform agents is a giant distraction. Agents are going to live everywhere: ServiceNow, Salesforce, ChatGPT, your IDE, your harness, wherever. It's a good thing platforms are building their own agents. It's good you're using your own agents. There won't be one winner.

Andy Berman avatar
Andy Berman@berman66·Nov 16

Long MCP.

I came to the conclusion that most SaaS should become an MCP server. we’re getting the economics of AI software backwards. A lot of AI startups are building their own chat interfaces, agents, and workflows, essentially reselling tokens from the big labs at low to negative gross margins. What if SaaS companies did the opposite? Take HR software. Instead of building an AI assistant, become the best MCP server for HR. Expose tools to: - Benchmark salaries against proprietary market data - Run compensation reviews and calculate pay equity - Model salary increases against a department’s budget - Recommend promotions based on company policies - Update employee compensation in the HRIS - Manage approvals, permissions, and audit trails Let ChatGPT, Claude, or Copilot handle the conversation, reasoning, and orchestration. Let the customer pay for the tokens. The SaaS company keeps the valuable stuff: proprietary data, integrations, business logic, permissions, compliance, and the system of record. You still charge a SaaS subscription, but you’re no longer paying to generate every answer. You can ship faster, with fewer engineers, and potentially restore software-like gross margins. The next generation of SaaS might not have much of a UI at all. Just an incredibly good MCP server.

Andy Berman avatar
Andy Berman@berman66·Nov 16

Some news: @Runlayer is a design partner of Personal Agent Protocol, alongside companies like Meta, Stripe, OpenAI, and other industry-leaders. This is one of those "Why doesn't this exist yet?" protocols, and it was a no-brainer for us to contribute. Today, personal agents have two bad ways of interacting with business systems. They use human credentials and browse human-first websites poorly, or can't work with the system of all, because the business blocks agents. Personal Agent Protocol (Poppy) fixes this for both parties. Companies can use the protocol to define what agents can do, which permissions they need for which actions, and where a human needs to step in. Humans can similarly define what actions their agent can take on their behalf. Their identity never gets exposed until they give explicit permission. It's secure, granular, and allows governance you couldn't get before. Proud to be a part of this protocol.

Post attachment
Andy Berman avatar
Andy Berman@berman66·Nov 13

In the past four days, four mid-sized companies have separately told me they vibe-coded their own version of software they were paying for. One asked why they were paying $100K a year for it. Pretty much every software player around the system-of-record is fair game. People are completely underestimating this shift over the next 10 years.

Andy Berman avatar
Andy Berman@berman66·Nov 13

A VP of IT told us "I can't tell my CFO how much the company is going to spend this month" He's stuck. Cap AI usage and your top 1% of users can't do their jobs. Lift the caps and finance comes after you. Until he can attach a dollar amount of ROI to every token used (positive or negative) he'll either be making finance or engineering angry.

Andy Berman avatar
Andy Berman@berman66·Nov 11

A public company told us they went from zero MCPs to 130+ in a year. Their CIO's biggest new problem is agents either borrow the user's permissions, or run on a service account. "None of which is ideal." Or the right way to do it. If your agent has your permissions, it's usually too much for what the task actually needs. Any prompt injection has the user's full blast radius. If it has a service account, it's typically shared between several agents, and you can't tell which agent did what. Every single one of our agents gets its own agent account: a Runlayer-native identity with its own Client ID/secret, policies, delegations, tokens and audit trail. I wouldn't run production workloads any other way.

Andy Berman avatar
Andy Berman@berman66·Nov 11

We were just named to The Information's 50 Most Promising Startups. They call it "The 50 startups most likely to become tomorrow’s giants." Based on their track record, they're usually right. This same list picked Cursor at a $400 million valuation, Polymarket at $200 million, and Grafana at $325M. Later today, at 12PM PT, I'll be speaking live for a conversation with The Information team and a few other members on the list. I'll drop the link below. Congrats to everyone else who made it, and thanks to @aaronpholmes and @TheInformation for the recognition. Most of all, thanks to the team. I'm proud of the work they've done to get us here, and land us on lists like this. On that note, this is your last chance to join us before we go parabolic. We're hiring for a ton of roles to scale enterprise AI adoption to every company on the planet. Join us: runlayer.com/jobs

Post attachment
Andy Berman avatar
Andy Berman@berman66·Nov 11

Runlayer Labs ran this attack on the Grok Bot vulnerability against our own agents, and found @Runlayer detects and blocks it. As soon as we heard about this attack from Profero's reporting, our team ran the sequence on our own agents. I was confident. Runlayer detects the attack where it starts, at the hidden instruction in the page the agent reads. I can confirm three separate scanners independently flagged it as prompt injection. We caught the moment the agent was hijacked. Runlayer flagged the agent dropping the user’s task to follow the hidden instruction, even when the agent described what it was doing in harmless-sounding terms. And when the agent tried to send secrets to the attacker’s server, Runlayer flagged the command before it ran. Normal installs, API calls and Docker commands didn’t set off a single alert. Grok's sandbox-escape path doesn't exist on Runlayer's own agent runtime, but architecture alone isn't the answer. It's clear organizations need true defense-in-depth, and they can't rely on the agent's architecture to save them. AI agents aren't capable of governing or securing themselves. Full analysis: runlayer.com/blog/grok-bot-v… If anyone at @spacexai is reading this, I'd be happy to chat more about how you can stop attacks like these. x.com/GelosSnake/status/2119…

We tried very hard to reach @bot ans @spacexai security team for 75 days. As no replay from them and no observable fix to the issue we are releasing the full disclosure today:

Andy Berman avatar
Andy Berman@berman66·Nov 10

"Some of it was pretty cool. Some of it was terrifying." A public company told us this after dozens of employees started building Claude dashboards on their data warehouse, under their own permissions, and sharing them company-wide. I honestly don't care that it's a security nightmare. You can fix those. It's much harder to fix "my workforce doesn't want to use AI" and clearly that's not the problem here.

Andy Berman avatar
Andy Berman@berman66·Nov 10

The secret's out, we're growing faster than ever. Join us before you regret it. We're hiring: runlayer.com/jobs Thanks for the shout, @lennysan

Post attachment

The best time to join a rocketship is before everyone knows about it. I’m excited to launch The Next 50: the 50 hottest early-stage startups to keep an eye on (and try to join). Using @harmonic_ai’s data, the juiciest startup dataset out there, we looked at every early-stage startup and ranked them on headcount growth, recent funding, and investor interest. We then picked the top 50. Here's the full list: lennysjobs.com/next50 We’ll update it quarterly, and I’d love your feedback on what other factors we should consider. P.S. This is in addition to the Lenny 100, the top 100 well-known companies with the highest talent density, tackling the most ambitious problems: lennysjobs.com/lenny100

Andy Berman avatar
Andy Berman@berman66·Nov 10

Our founding engineer built an MCP explainer and it hit #2 on Hacker News. Somehow he survived the comments, too. Everyone was talking about how the hard part of MCP is auth, and it only really pays off inside the enterprise. Agreed, and that's the part we work on. You should check it out. x.com/rafalwilinski/status/2…

most of us (definitely me) thought MCP was just about tools working with it for the past +1 year taught me there’s so much more: resources, OAuth, sampling, elicitation, tasks, etc so i built a visual explainer. Every message on the wire, step by step: mcp-explained.com

0:15