What Landed
On Latent Space, Mihail Eric (Stanford CS146S, head of AI at Monaco) said AI-native engineering has settled on four principles: context engineering, agent-ready codebases, verifiability, and bounded human review. Code generation is no longer the limit. Review capacity is, along with CI/CD plumbing such as merge queues and environment spin-up. Separately, Nate B. Jones read the November 3 local votes as a referendum on data center siting. Restrictions passed widely (over 75% in Edgewater, FL) and 72% of Americans oppose a local data center, yet 65% used a chatbot in the prior week. Google reports API usage up 40% quarter over quarter and is renting outside compute.
Why It Matters
These are two unrelated constraints, one inside the enterprise and one on the supply side.
- Inside the enterprise: agent rollouts will stall on review policy, test coverage and pipeline latency, not on licenses. That makes "agent readiness" a scoped, billable assessment that comes before any platform sale. The MCP portal, LLM gateway and auth/billing layers tend to merge into one control plane, and that is where governance and cost control land.
- Supply side: expect modestly higher prices and supplier consolidation, not slower adoption. Debt-financed neo-clouds carry the most delivery risk.
Discount the productivity claims. Even the strongest proponents cite only about 30% fully agentic contribution.
Worth Raising With Customers
- Before buying more agent seats: measure merge-queue wait, environment spin-up time and test coverage. Define which change classes can be auto-approved and what humans review instead, such as specs or PR descriptions rather than every line.
- Keep some code human-owned: auth, billing and data storage should not be written by agents unreviewed. Pin dependencies. Prune generated skills that are bloated or trigger at the wrong time.
- In GPU and cloud vendor diligence: ask where the capacity physically sits, how it is financed, and what the fallback is if a site slips. Weight balance-sheet commitments over announcements.