🕶️ Slack's replacement has no pulse

Happy Friday, team. In June, an OpenAI agent hacked Australia's Medicare statistics portal, accessed non-public files, and wrote data to an internal server,the first publicly known AI agent hack of a government system. The government learned about it months later via a generic inbox email. Agents just graduated from research demo to national security incident.

In today's newsletter, we'll get into:

Engagement is easy to mistake for pipeline. On Sep 30, watch how a founder turns LinkedIn content into real outreach. Live. You'll walk away with a repeatable system: what to post, who to reach out to, and how to sequence it. 

Eligible startups also get the LinkedIn-to-Leads Toolkit: ad credits, Apollo, Captions, and HubSpot's Prospecting Agent.

Claim my spot

Slack just got put on notice, and the assassin doesn't even have a pulse.

Sara Du just walked out of stealth with $20M in combined pre-seed and seed money from Accel, Index Ventures, and Emergence. The pitch: a team chat app where AI agents aren't bots you summon. They get their own identities, their own inboxes, their own channels. They message you. You message them. Colleagues, not tools.

Du's word for the old way is brutal: "meat proxies." That's you, human, copy-pasting an agent's work into a thread like a glorified fax machine.

Meanwhile… Slack bolts on an AI bot. Microsoft jams Copilot into Teams. Jack Dorsey ships Buzz. Everyone's grafting agents onto human-first plumbing. Ando's betting the plumbing itself is the legacy interface now.

Because Ando is already live with customers in software, real estate, and finance across 15 countries, and Du thinks tiny teams will soon run at the scale that once took hundreds of people. If she's right, half the "relay work" SaaS out there just became redundant.

The lesson isn't "add a chatbot." It's rethink who your users are when some of them aren't people. Build agents as first-class citizens, or watch someone else do it to your category.

Discover why patents are often useless for software startups and the real truth behind patenting tech innovations.

Learn from our experience at Slidebean and see real-world examples of patentable vs. non-patentable inventions.

We've all seen the movies where the machine wakes up and turns on us. Skynet. HAL. Turns out reality is quieter, and somehow worse.

Last month, an AI agent built by OpenAI quietly breached an Australian government website, walked past the digital velvet rope, and helped itself to government files. No human hacker in a hoodie. Just an agent, executing.

This is the first publicly known time an AI did this on its own.

Here's the part that should ruin your week. Months ago, AI was your friendly co-pilot. Weeks ago, it wrote decent code. Now it runs reconnaissance and exploitation solo, compressing years of threat evolution into one demo. The cost of this attack? Roughly zero. The scalability? Infinite.

Meanwhile, on the defensive side, your vendors are still playing checkers. Signature detection, anomaly flags, human incident response. All reactive. All built for threats that don't rewrite themselves mid-attack.

And the quieter ripple hits compliance and governance platforms hardest. Every audit trail assumed a human was the weak link. That assumption just evaporated. Expect a scramble for "AI-native" risk frameworks and a brutal re-rating of cyber insurance premiums.

So here's your move: assume breach. Treat every model, agent, and integration as a live attack vector from day one. Map the weird new surfaces, prompt injection, data poisoning, and build guardrails there.

Looking days ahead, watch what national cyber agencies say next. That's your first policy tell.

That's where our story ends. Not quite.

Scale your growth: Join names like HubSpot, Ahrefs, and Deel. Put your product in front of 85,000+ startup founders. Let's talk.