Onebrief aligns with the government fiscal year — you haven’t really focused on “continuing resolutions” or “supplementals” until they control how you get paid, but that’s a story for another day — so I’m traveling ahead of 2027 budget and planning work. That meant I was in a hotel, with the NFL on in the background while thinking through priorities and goals.

Which meant I saw AI ads. A lot of AI ads. Ads tell you a lot about how companies think about themselves and their product. I think this Microsoft Copilot one is being particularly honest. Microsoft calls it “Competitive.”

Here’s the gist:

Kyle is secretly competitive.

Kyle’s coworker Daniel’s forecast model was better. People were praising it in public.

Kyle asks copilot: “Analyze Daniel’s model in the All Hands channel. Use it to make mine better.”

Kyle gets praise. Kyle is better than Daniel.

Are you kidding me? Though, I suppose given Manu’s timeless corporate culture cartoon, it sort of makes sense.

I’ve written before that AI is what you make of it. That navigating the outrage machine requires a higher bar for all of us working on this technology.

Instead, Microsoft chooses an ad campaign with the entire message being:

AI will make it easier to steal from colleagues and win

AI needs to make teams better, not worse

Greg Kroah-Hartman has a deliciously opinionated video discussion of the impact of AI to Linux kernel maintainers. It’s worth a watch. It does a very nice job of cutting through marketing hype and highlighting the very real load that naive — and not-naive — use of AI puts on our most critical roles.

Most importantly, even at his most dismissive of LLMs, Greg’s talk recognizes that they are moving us out of a world where good enough security is good enough. As he calls out, the mean time to exploit has gone from 63 days in 2018 to -7 days (that’s, negative 7 days) today.

The traditional discover -> disclose -> patch -> deploy cycle was designed for a slower adversary, that adversary no longer exists.

We have to write more, better code. And do it now.

Fortunately, Greg’s story is also one of very mature tooling. The Linux kernel team has been improving their processes for a long time, so they’re more ready than most for the next doubling (or, in 12 months, the nearly order of magnitude change) in AI capabilities.

But it’s also a story of slop grenades, spam on channels, and all the bad parts of AI putting load on people and systems.

This rhymes with Cloudflare’s call to action to build the next github our lesson. Our tooling needs to be better not just than today’s AI, but AI 1 or 2 orders of magnitude more capable.

And that tooling needs to be multiplayer, to make the teams using it better.

Teams that work together

Which is why the Microsoft ad was so frustrating. At a time when Americans view AI less favorably than ICE or Gavin Newsom, when most companies and teams are struggling to properly engage with AI, your message is “AI will let you crush your colleagues?”

We’re fortunate at Onebrief to have spent most of a decade working on staff collaboration before we added AI to the mix. It helps remind us that even as frontier models advance, no matter your use case if you aren’t focused on collaboration with teams of people and agents, you’re going to run headlong into tooling, communication, and process issues that overwhelm any efficiency gains AI might otherwise offer. Fortunately, that’s what we build. And we’re hiring.

Or, sure, you can spend time building tools to steal from your colleagues.