"With AI we can fire all our vendors and build everything ourselves." That's the hottest take in enterprise IT right now. But I've heard this story before.
The pattern repeats every decade. Outsourcing: cheaper, faster, not our problem. Then insourcing: strategic, we need control. Then back again. Agile was going to replace waterfall forever. Cloud was going to make on-prem extinct. Each time the pendulum swings too far, then corrects.
And we're at the start of a new swing.
A seductive story
The narrative is tempting: AI makes developers 10x more productive, we'll rewrite legacy systems in a year, we don't need ERP vendors, CRM vendors or systems integrators. Just hire two AI engineers and ship everything ourselves.
I'd bet against it, and here's why.
The slowness of large enterprises doesn't come from technology. It comes from the organization. And that doesn't change because Claude can produce a lot of code.
Risk management, compliance, security review, change control, stakeholder alignment. All of this exists because large companies have a lot to lose. That calculus doesn't change just because a model can generate code.
The unglamorous work nobody started
AI won't fix organizational complexity until companies redesign their processes one by one: clarify the standards, define what's allowed and what isn't, and encode those decisions into agents that execute faster than people. That's deep, unglamorous work. Most companies haven't even started it.
My prediction: in 12 to 18 months we'll see the swing back. The arguments for a healthy mix of internal capability and external partners didn't disappear. They're still there. They'll become visible again once the first wave of "we'll build everything ourselves" projects hits the wall of enterprise reality.
The pendulum always finds its center. The winners won't be the ones who swung furthest, but the ones who found the balance faster.
If you're weighing where in-house AI development makes sense and where a partner pays off, get in touch. I'm happy to help you find that balance before trial and error does.
