A $30M startup wants to watch every AI agent in your company

In this edition:

• This week: Claude moved into Slack channels, Gong shipped a governed-agent layer for revenue teams, and a $100M buyer disclosed swapping to a 23x cheaper model

• Under the radar: Runlayer raised $30M to be the neutral, cross-provider watch-layer for every AI agent in your company

• What's on the calendar: GPT-5.6's rumored launch, the OpenAI IPO roadshow, and Anthropic's Fable 5 restoration after the Glasswing disclosure

THE WEEK IN AI

THE WEEK IN ONE SENTENCE

Claude moved into Slack, Gong shipped a governed-agent layer for revenue teams, and the gap between "which model is best" and "which model is cheapest for this workload" closed enough that a $100-million buyer named it publicly, while the layer that watches and prices every agent in the company got funded as its own category.

THREE SIGNALS

01 • Agents

Anthropic moved Claude into Slack channels, and Gong did the same for revenue teams

Anthropic launched Claude Tag on Wednesday in beta for Enterprise and Team customers. The integration replaces the old Claude in the Slack app with one shared Claude per channel that remembers context, runs tasks asynchronously, and works in front of the whole team. An administrator pairs Claude with a workspace, grants access to specific tools and data sources, sets a budget cap, and defines which channels Claude can operate in. Anyone in those channels can tag @Claude and delegate work.

The day after, Gong announced Mission Big Dipper and the Gong Revenue Harness, an agentic execution layer that lets RevOps leaders and sales managers build governed AI agents on top of Gong's call and account data without engineering support. The framing is explicit in Gong's release: generic tools "break in production because they lack revenue context, governance, and human-in-the-loop control." The product is a governed-agent layer for the GTM team.

What I keep hearing in client conversations this quarter is that the AI surface is moving. Two years of enterprise AI lived inside a browser tab or a sidebar in an IDE. This week, two of the biggest vendors in their categories announced the next surface: the agent lives in the room where the team already coordinates. For the operator, the question stops being "which seats do we license" and becomes "which channels does this agent sit in, what tools does it hold credentials for, and what is the cap." Anthropic and Gong both put those decisions in the administrator's hands by default. The work for the operator is making sure the channels in scope match the data-access posture the security team would write down if asked.

AI READY PRO

I will get you AI trained in 30 days for free

After several months of running it quietly with top AI operators and teams, I’m excited to launch AI Ready Pro today to our newsletter subscribers like you.

It’s a 30-day personalized AI training program that is personalized to you and where you are in your AI journey. It’s the culmination of 100’s of hours spent teaching AI to top Fortune 500 AI teams and operators (and working with Mark Cuban).

Each day, you receive a 10-15 minute exercise to complete to improve your AI skills.

This isn’t vague AI theory or a way to pitch you a tool. It’s 30 days of learning to use AI in your real work. So you come out the other side with the skills to use AI to actually improve you and your teams output.

If you lead a team, AI Ready Team is open today too, also for readers first. It’s the same engine, but to train your whole organization. You’ll get a detailed view of:

• Your team’s AI readiness level + detailed strengths & weaknesses

• Ranked list of AI automation opportunities

• Get tactical advice on how to advance AI efforts in your org

Leading a team? Take the Team assessment instead.

02 • Model

The cheaper-model swap got its first $100 million disclosure

Ensemble Health Partners, a vendor that processes hospital revenue cycles, told The Information on Tuesday that it is on track to spend up to $100 million on AI this year and has had "success" moving workloads from OpenAI's frontier tier to a model 23 times cheaper from the same vendor. The reporting paired the disclosure with an OpenRouter data point: 65% of tokens processed through the developer-routing platform in June came from open-source models.

That changes the procurement conversation. When the question was "are we on the best model," it could only be answered by trying every vendor. The new question is closer to "are we on the cheapest model that does the work," and that question has a measurable answer per workload. Ensemble is the first $100-million-class buyer to put a name on it publicly. The work is unglamorous, reading prompts, running the smaller model against the same evaluation set, confirming the output is within tolerance, but the companies that do it are quietly compounding savings against the ones that don't. The honest hedge: this only matters where the workload is high-volume and the evaluation criteria are clear. Frontier still wins on complex reasoning, agent loops, and anything customer-facing without a safety net.

03 • Security

Four frontier-security vendors drew the boundary around the agent

In one week, Microsoft, Google DeepMind, CrowdStrike, and Tigera all published security architectures for AI agents. Microsoft's research team disclosed AutoJack, a three-vulnerability exploit chain in AutoGen Studio that lets a single malicious web page execute code on a developer workstation through an AI browsing agent. Google DeepMind put out its first AI Control Roadmap, comparing the right architecture to a driving instructor with dual controls. CrowdStrike published Continuous Identity for AI Agents. Tigera shipped an eBPF runtime that fences agent containers from the host.

Four publications, same week, same posture: the AI agent inside the stack should be treated as an entity that will at some point try to do something the operator did not ask for. The week the four arrived is the week agent-security stopped being a research category and started being a procurement line item. By Q3, the security team is going to be asked about it on the renewal currently in front of operators, and the contract written 12 months ago did not anticipate the question.

UNDER THE RADAR

Runlayer raised a $30 million Series A on Wednesday led by Felicis, with Khosla Ventures participating. The startup, which launched out of stealth seven months ago, builds the corporate app store and control room for AI agents. Employees plug in any AI tool (ChatGPT, Claude, Agentforce, custom agents) and get a pre-approved way to use them, with company data already connected and guardrails set. IT and security get a single dashboard showing what every agent is doing, what data it touched, and how much it cost. The customer list as of this week includes Instacart, Gusto, Opendoor, dbt Labs, AngelList, and a Fortune 500 bank running Runlayer across 100,000 employees and 200,000 devices.

The line worth carrying is from Felicis GP Jake Storm, who led both rounds. "This is a Switzerland business. No platform can own this. A neutral, cross-provider control layer is absolutely critical if we actually believe in the future of agents performing work."

That is the procurement category forming in real time. Microsoft will want to own this seat from the platform layer. Salesforce will want to own it from the data layer. OpenAI and Anthropic will both want to own it from the model layer. Runlayer's pitch is that the buyer who actually carries the agent inventory does not want the layer that watches the agents to be owned by any of them. The week the agents moved into the rooms where work happens is the week the watch-layer got funded as a category, not a feature.

QUOTE OF THE WEEK

"Who watches the watchman? That's what we do"

SPONSORED BY CLUTCH

Hire secure AI teammates that work 24/7.

Hire pre-built AI teammates. Give your engineers and operators a platform to ship their own AI apps. Stop losing sleep about what is running where.

Clutch is the platform behind both: pre-built agents for the workflows your ops team should automate first, plus the integration plane your team's vibe-coded apps and Claude Code projects plug into. One platform. Real production. Visible and safe by default.

Built for ops, engineering, and security teams that are tired of the shadow-AI surface area inside their own company.

WHAT’S ON THE CALENDAR

• GPT-5.6 is expected to land as soon as next week. Polymarket odds sit at 88% on a late-June or early-July release, codenamed Ember Alpha. The OpenAI IPO roadshow is the macro pacing item underneath it — the $1 trillion-plus filing is now public, and any S-1-stage disclosures move on a fresh tempo.

• Anthropic's Mythos and Fable 5 restoration is the third item to watch. Following the Project Glasswing classified-systems disclosure, the same model the US government briefly restricted is now being run inside classified environments to find defense vulnerabilities. The policy posture is still moving.

Have a good weekend,
Haroon