Tal Peretz

Tal Peretz

Founder @Runlayer

The MCP vs. Skills debate is over. Skills are now part of MCP. Our customers have been saying this for months: Skills are coding a behavior that can change an agent’s trajectory. They shouldn’t be distributed as ungoverned markdown files. They need the same control plane as tools: → Who wrote it? → Who approved it? → Which version is running? → What can it access? → What did it do? MCP just made skills discoverable. Now enterprises need to make them governable. Let's go.

Tal Peretz

Founder @Runlayer

We raised $30M so you can go all in on AI. Every AI champion and security leader we met asked us the same questions: • How do we say yes to agents instead of no? • How do we connect them to our real tools, data, and context? • How do we get visibility and control everywhere an agent touches? That’s why we built Runlayer. The golden path for AI. So an employee can hand real work to any agent, with any tool — and security can stop the bad guys in real time. Enablement and control in one platform. The future isn't one model or one agent. It's millions of agents, tools, and workflows running across every surface of the company. That world needs one layer that lets all of it run fast, and safely. This is the path to ‘AI Native’: → Discovery: shadow agents, MCPs, and skills found across web, cloud, and endpoints → Approvals: only reviewed, scanned, approved tools ever attach → Identity: every agent action traces back to a real, managed identity → Authorization: least privilege, scoped per tool, default deny → Context: the right tools, skills, and knowledge, delivered to your agent → Runtime: high-risk actions intercepted, scanned in real time for intent drift, with an instant kill switch → Observability: audits for every call, across any client, streamed to your SIEM → Governance: static and dynamic policies, agent IAM, one control plane For builders, you don't need to worry about these, agents become easy to connect, run, and improve from any client. For security and platform teams, you finally enable the rollout instead of slowing it down. Teams that came to us were done waiting. Some had walked the RSA floor hunting for a control plane to sit between their agent harnesses and their hundreds of SaaS and internal apps. Others had started building it themselves — and realized they needed an army. Today, Instacart, Gusto, Decagon, Opendoor, dbt Labs, AngelList, Lemonade, and multiple Fortune 500s run on Runlayer. We'll keep going wider and deeper, until putting agents on your core business isn't a leap of faith. It's just how you work.

Tal Peretz

Founder @Runlayer

A teammate cloned me this week. Not metaphorically. He pointed an agent at 90 days of my Slack history, distilled the patterns, and wrote a skill titled something like "You are Tal." Tone. Anti-patterns. How I think out loud while typing fast. All of it, packaged as a reusable instruction file any agent in our stack can load. We laughed. Then we kept going. What if every person on the team has a skill? What if the skills have access to Slack, GitHub, Linear, and company cards? What if we spin them up in a separate workspace and let them run - bizarro-Runlayer, a parallel org of digital twins, working 24/7 while we sleep? Here's the reframe. We've been talking about skills as if they're reusable prompts. They're not. A skill is an identity container. The moment you can compress a person into one - voice, judgment, defaults, failure modes. You can compress a team. Then an org. The unit of composition in an AI-native company isn't the employee anymore. It's the skill. Employees are the source material, skills are the deployable artifact. Dynamic auth makes them portable. Every agent you run can now wear a different person's judgment the way a function call wears a different signature. The org chart just went multiversal.

Tal Peretz

Founder @Runlayer

Two 'competitors' copied our website this month. Layout, messaging, positioning - close enough that our lawyer said we should send a cease and desist. We decided not to. Not because we didn't care. Because we believe it's a waste of time. The time and energy you spend in legal disputes is time and energy you don't spend shipping. So we chose the other path. Outbuild them. Outship them. Make the copied version outdated. When someone copies your work, the instinct is to fight. It feels personal. It is personal. But startups don't win by protecting what they built yesterday. They win by making yesterday's work obsolete with what they ship tomorrow. The best legal strategy at a startup is shipping faster.

Tal Peretz

Founder @Runlayer

Anthropic launched MCP Apps yesterday. Most teams will treat MCP Apps as a new surface to secure. That's the wrong mental model. Apps like Asana, Figma, Slack, and Box now render interactive UIs directly inside Claude. It looks new. But under the hood, MCP Apps are standard MCP servers. Same protocol. Same traffic patterns. Same security surface. That's the point we've been making about protocol-level governance: when you secure MCP at the transport layer, you don't play catch-up every time the ecosystem ships something new. Runlayer customers have full coverage on MCP Apps today. Tool definitions, tool calls, resource fetches, OAuth tokens - all of it flows through the same security controls you've already configured. Same visibility. Same audit logs. Nothing to update. This is what we mean by Enterprise MCP infrastructure. The protocol evolves. Your governance keeps pace automatically.

Tal Peretz

@talperetz_

Your team is running MCP servers you don’t know about. Different clients, different devices, no central view. That’s a security problem. Our latest release fixes that. Runlayer 1.25 gives you full visibility into what’s actually running, and the controls to lock it down ↓

Tal Peretz

@talperetz_

Today we're announcing a partnership with Cursor on Cursor Hooks. Hooks let you extend your agents with custom scripts. The Runlayer Hook gives you full visibility, control and security for every MCP call. Engineers just keep shipping.

Tal Peretz

Founder @Runlayer

We launched Zapier MCP 1.0 to help LLMs take real action. It worked—but it was just the beginning. Today, we’re rolling out Zapier MCP 2.0—completely rebuilt from the ground up. We worked closely with the team at Anthropic to deeply integrate Claude and help bring real value: 🎯 Native support for Claude, Cursor, and Windsurf 🔌 Works with any of Zapier's 8,000 apps—even private ones 🛡️ Secure OAuth2 support 🌐 Multi-server support 🧠 Advanced parameter selection for AI actions If you want to use AI that does stuff—not just says stuff—this is for you.