Everyone’s building their own Company OS

Hiya 👋

Three stories landed this week that read like chapters of the same book.

The biggest law firm in the world just decided that buying AI off the shelf isn't enough. Goldman quietly published a number that says the labor market is already shifting. And a survey landed that puts a percentage on what separates the companies getting value from AI from the ones who aren't.

Let's get into it 👇

One of the world's largest law firms just told the Financial Times they're setting aside $500M to build custom work orchestration software and AI platform instead of relying on the same off-the-shelf tools their competitors use. They intend to spend hundreds of millions more over the next few years.

What's the story? Jon Ballis, Kirkland's chair, said the goal is to "take the collective intelligence of our institution and deploy that throughout our firm." Translation: the moat isn't the AI. The moat is what their lawyers know, how they work, and the institutional knowledge that took 50 years to build. Off-the-shelf tools give competitors the same capability. Custom software trained on your own work doesn't.

This is a signal amongst the AI hype. CEOs, partners, and operators across industries are saying the same thing: any company that mostly works with ideas, documents, and institutional knowledge should be asking why they aren't doing this. The economics of building have collapsed. The cost of being indistinguishable from your competition hasn't.

We saw this writing on the wall and our new site reflects what we’ve come to know and Kirkland is showing publicly: the companies pulling ahead are building software customized to how they work and they weave AI into those workflows.

Our new site went live recently and reflects how that bet became the standard for how we work with clients…and what they want.

Go deeper:

🔗 Read the FT story → Kirkland & Ellis sets aside $500M for its own AI platform 

🔗 Check out our new site → withswitchboard.com

The WSJ ran a piece last week with a headline I didn't expect to see this soon: companies have a new AI problem, and it's too many agents.

What's the story? The numbers are real. FICO has 3,500 employees building dozens of new agents every day. DaVita employees have created over 10,000 internal agents. GitLab and Lyft are now navigating what one CIO called "agent sprawl." The same dynamic showing up at companies with $100B in market cap is starting to show up at small and mid-market companies too. Anyone can spin up an agent now. Most do it without telling IT.

This was inevitable as execs told everyone to “start using AI” and governance didn't catch up.

So, you end up with multiple agents doing the same task in different ways, agents calling other agents, costs creeping into computing bills nobody is tracking, and security teams discovering things they didn't know existed. The fix isn't preventing employees from building. That ship sailed. The fix is governance and connective tissue. Knowing what exists, what it touches, and what owns what.

If you run a small or mid-market company: The instinct here is to lock things down. Don't. The companies handling this best are doing the opposite: making it easier for non-engineers to build, while putting governance around what gets built. A registry of what exists, owners for each agent, fallback plans for critical flows. Boring infrastructure that lets the interesting work happen safely.

Go deeper:

📰 Read → Companies have a new AI problem: too many agents

Goldman published research last month estimating AI is currently a net drag on the U.S. labor market: roughly 16,000 fewer jobs per month over the past year. Substitution wiped out about 25,000 jobs a month. Augmentation added back about 9,000. The pain is concentrated in entry-level and Gen Z roles. The NYT picked it up this week with an opinion piece arguing this is no longer a future scenario. It's already showing up in the data.

What's the story? The interesting part isn't the headline number. It's the breakdown. AI doesn't replace whole jobs at the macro level yet. It replaces some tasks and augments others. The net result so far is modest. But the distribution is brutal. Younger workers in knowledge work are seeing hiring slow noticeably. Companies aren't announcing layoffs. They're just not opening as many seats.

And then this Sunday, Pope Leo XIV published his first encyclical, Magnifica Humanitas. Forty thousand words, focused entirely on AI and human dignity. He presented it alongside Christopher Olah, an Anthropic co-founder, who said this from the podium: "There is a real possibility that AI will displace human labour at very large scale. If that happens, supporting those displaced will be a moral imperative of historic proportions."

It's worth pausing on the unusualness of that sentence. The Pope and an AI lab co-founder, on the same stage, telling each other that this is real and the responsibility is shared. Goldman's CEO David Solomon has been making a more optimistic case: that AI cuts task time, but markets respond by raising the bar. Analysts who built one chart now produce sharper modeling, broader analysis, faster turnarounds. The work expands. Both can be true. Tasks get automated. Expectations rise. Some people gain. Others get squeezed out before they ever get a chance to start.

If you run a small or mid-market company: Don't outsource this question to the macro debate. Look at your own org. Where are you about to hire and considering not to? Where are you about to lay off because AI handles the task? Whatever you decide, decide it deliberately. The companies that come out of this period with reputations intact will be the ones that thought about it as a leadership question, not a productivity question.

Go deeper:

📰 NYT opinion on the Goldman data → The AI job crisis is here

⛪ Pope Leo's encyclical and Anthropic → Why is an AI company helping launch the Pope's AI encyclical?

Harvard Business Review Analytic Services and Appian surveyed 385 businesses already using AI. One number popped out and is worth committing to memory.

What's the story? 71% of organizations that embed AI directly into their processes report moderate or substantial value. Among the rest of the sample, that number is 16%. Almost 5x difference, driven by one variable: whether AI is built into the workflow or sitting next to it. 69% of respondents said legacy systems are limiting their ability to scale AI. The conclusion the report came to: getting real value from AI requires three things happening at once. Modernizing core systems. Integrating fragmented data. Redesigning workflows with AI embedded from the start. Not bolted on after.

This lines up with what we keep seeing on the ground. The companies getting real value from AI aren't the ones who bought the most tools. They're the ones who treated AI as an operating decision, not a software decision. Workflow first. AI second.

If you run a small or mid-market company: Trying to redesign every workflow at once is how most of these projects stall. Pick one. The one where the cost of being wrong is high and the cost of being slow is higher. Start there. The 71% in the survey didn't get there by buying AI. They got there by changing how work happened around it.

Go deeper:

📊 Read → High-value AI use cases lag behind enterprise hype

Know someone who’d enjoy this newsletter?
Forward it along, or send us a note if there’s a topic you’d like us to unpack next.