Boring infrastructure wins
Deleting a Redis dependency moved adoption further than any feature announcement this month. That is not a joke about databases.
A stateful dependency is a deployment veto
Redis means another service to run, secure, monitor and pay for. It converts something that could have been a container behind a load balancer into an operational commitment with an owner and an on-call rotation.
Removing it is not a feature. It is the removal of an excuse.
And the curve shows it — a billion cumulative SDK downloads, nearly 500 million monthly. Half a billion a month against a billion lifetime means the growth is recent and steep, not a long tail.
What it cost
Sessions were there for a reason, and statelessness usually buys deployment at the price of something — either clients carry more context per request, or features that depended on server-side continuity get reworked. Adoption at this pace suggests the trade went the right way, but it was a trade.
The standard emerged where it had to
No consortium, no committee, two years. The integration layer standardised because no vendor could defend it alone — and the proprietary alternative is being switched off by its own publisher this month. Meanwhile the models converge on nothing.
The bill arrives later
A very large ecosystem of third-party servers holding tool-calling authority, with no review layer between registry and production. Package ecosystems learned this expensively; this one is younger and wired directly to actions. Which is why enforcement points are moving in front of the model.
What this asks of you
Audit which servers you actually trust and write down what each is permitted to do. And when you evaluate any tool this quarter — including agents reporting 44% better accuracy on 48% fewer tokens — ask what it needs you to operate, not what it can do. That is the number that decides whether it ships.
Model Context Protocol — The 2026-07-28 Specification → · The New Stack — Atlassian wants developers to finally like Jira →