I had dinner with a group of senior engineering leaders in Minneapolis last week, and the conversation didn't stay where I expected. We started on AI adoption in the SDLC. We ended on what "ready" actually means right now.

Here's what stuck with me.

Mythos Readiness Is Already Here

Most of the table hadn't fully mapped out what Mythos-class models mean for their own environment. That didn't stop the conversation. It kept circling back to the OpenAI compromise of Hugging Face, and less to the damage done to Hugging Face than to how the model got out of containment in the first place.

At one point I said it plainly: it's the new apex predator, and it doesn't sleep. Nobody at the table disagreed.

Resiliency Is Becoming a Cost Conversation

Resiliency strategy and execution matter more now than they did a year ago. The next real tension point isn't AI capability, it's cost versus resiliency, and security is going to be the one pushing that boundary.

My advice: let security help make the case. Robust resiliency for your most critical applications is a harder sell when engineering pushes for it alone. It's an easier one with security standing next to you.

Predictability Over Speed

As AI takes on more of the delivery pipeline, "faster" stops being the right measure of success. Delivery outcomes need to be predictable before anyone will trust them, and predictability doesn't happen by accident. Standardization came up again and again, not as a compliance box to check, but as the thing that actually drives consistency at scale.

Figure out your negotiables and non-negotiables for your SDLC before you're forced to decide under pressure.

The Part That's Actually Hard

Every technical thread we pulled on ran into the same wall: executives. Not because leadership disagrees with the direction, but because bringing them along on the pace of change, without losing their confidence, is genuinely hard. Nobody at the table had a clean answer, myself included. What we had was agreement that this is now core to the job, not a side conversation you get to after the technical strategy is settled.

That's probably the real takeaway. The hard part of AI transformation isn't the technology. It's the trust you keep rebuilding with the people above you while you build it.

Closing

Minneapolis didn't leave us with a finished playbook. It left us with sharper questions, a shared vocabulary for the tradeoffs ahead, and a room of peers working through the same thing at the same time. That's usually the point.