In partnership with
Observability you can act on, whether you build it or lead it, plus the few signals worth your time.
October changed the question. It is no longer who owns the thing you depend on. It is whether you can prove what it did. A control, a contract, eight hours on somebody else's cloud: every piece came back to that test.
|
The person who knew why is not evidence. Neither is a green panel. |
This month's deep dive
Proof of Control, or Proof Nothing Broke Yet. A control that works and a control nobody has ever exercised produce the same telemetry on a quiet day. Both look like nothing. The piece reads what the logging article under DORA actually asks for and ends with a one-minute test for telling the two apart.
And the podcast is back. Tech Tuesday 5, What Is an AI-SRE?, takes one label stretched across four very different things and gives you two questions to ask whoever is selling you one. Listen to the episode or below or watch it on YouTube.
METRICS & MAYHEM
What Is an AI-SRE? The Autonomous Ops Reality
On the blog this month
Two full pieces went up in October:
Proof of Control, or Proof Nothing Broke Yet. This month's deep dive, above.
Your Vendor Just Bought Your Other Vendor. The big vendors are buying what sits between your data and a decision, and the challengers are going after your bill. Four checks before your next renewal.
Where each of those went, and why you may not have seen them
A blog goes to the stream it was written for. The Signal goes to every one of you each Friday.
Here is where October's two landed.
Leadership stream, 2 pieces:
On the leadership stream, both reached you. On the technical stream, neither did; both are on the web if you want them. If you never picked a stream, you may have missed them; a few of you got the leadership pair because of the role you gave when you signed up. Every one of them is on the web right now, at the links above, and picking a stream below fixes it.
Which do you want more of?
What's trending this month
Last month I promised to come back to one argument: who owns the thing once the person who knew why has gone. In October it stopped being an argument and turned into paperwork. The FCA's PS26/2 and the PRA's PS7/26 take effect on 18 March 2027, and they ask for a register of material third-party arrangements, filed every year. Every line on it is a claim about something you depend on, and a claim needs evidence behind it. The person who knew why is not evidence. Neither is a green panel.
Then the month supplied the test case. On 20 October, a power event took down one of Google Cloud's three London zones from 06:08 UTC, and Lloyds, HSBC, Royal Mail, Rightmove and DPD went with it. Google's account ran to one sentence, and for anyone who did not keep their own telemetry, that sentence is the record. On 1 October, Dynatrace completed its purchase of Arize, one of at least a dozen observability deals this year by my count, so a contract somebody chose because the supplier was independent now sits with a platform vendor. Same shape both times: the thing you depend on most is the thing you control least, and somebody else writes the account of what it did.
What actually changed, rather than what got loudest: the question moved from who owns it to whether you can prove it, while the evidence is being switched off to save money. In Splunk's own survey of 2,050 IT professionals, reported by Network World below, 38 per cent have cut telemetry coverage or retention to control cost. A vendor's survey, so read the finding, not the conclusion. I will come back to it in November and tell you whether anybody has started keeping their own copy of the signal for the things they rent.
Good posts I saw this month
From another voice in the space, worth your time and credited to her, not me:
Jessica Kerr (Honeycomb): our limits are not our imagination. Her case is that AI can make code more rigorous than code written by hand, through practices like property-based testing and formal verification, but only for teams that redesign how they work rather than using it to speed up old habits. Set it beside this month's argument. If rigour is getting cheap, not having had time to test it is about to stop working as an answer.
Worth your attention
Three listens and three reads that earned it this month, no two from the same source:
[Podcast] Charity Majors with Ben Lorica on O'Reilly's Generative AI in the Real World, 15 October. AI makes code cheap and moves the bottleneck to understanding what got built. Listen for why she prefers owner on the loop to simply being in the loop: it asks for active understanding and control of what the agents are doing.
[Podcast] Sam Crowder, head of product at LangChain, on Schneider Electric's AI at Scale podcast, 26 October. A demo does not show how an agent performs at scale; production needs tracing and evaluation, feeding the next round of tests. He also lands on shadow AI, which the 9 October Signal argued a moratorium tends to produce.
[Podcast] James Ward, principal developer advocate at AWS, on The New Stack podcast, 28 October. The detail worth the listen: MCP carries no trace IDs of its own, so AWS follows tool calls by session identifier. If you will ever be asked to evidence what an agent did, that is the gap.
[Article] AI agents drive up telemetry costs, complicate network observability, Network World. Denise Dubie on Splunk's State of Observability 2026: 57 per cent say observability now costs more than the infrastructure it watches. Every voice quoted works for Splunk.
[Article] The case for OpenTelemetry maintainership, OpenTelemetry blog. Jack Berg, a maintainer and Technical Committee member, argues the project's constraint is maintainers, not ideas or adoption, and by his analysis more than three in four of them work for observability vendors. If OpenTelemetry is your way out at renewal, know who maintains the road.
[Article] Google Cloud says a 'power event' caused today's outage, The Register. Paul Kunert's timeline of the London zone failure, from 06:08 UTC to most customers back by 14:03. Keep it beside your own records of that morning, if you have any.
Next month: what an agent should be allowed to do on its own, starting with the AI-SRE deep dive on the blog on Tuesday 3 November. Before then, draw the ownership map from the renewal piece: every telemetry vendor you pay, who owns it today and who owned it eighteen months ago. One sheet of paper. If there is a surprise on it, you have found November's work.
No Signal last Friday. That was deliberate. This Digest is the week's email from me, and I would rather send you one than jam your inbox with two. The Signal is back on Friday 6 November.
Allan
📕 The book is out. Metrics & Mayhem: A CTO's Guide to Observability That Actually Works. Kindle is live now; paperback and hardback launched on 1 June.
You built the audience. Can you reach them?
You can have thousands of people following your work and still need an algorithm’s permission to reach them. Creators know the drill: you post, you pray, you check the numbers, you try again tomorrow.
A newsletter changes that relationship. On beehiiv, you can build a subscriber list you own, reach readers directly, and turn their attention into a sustainable business with subscriptions, sponsors, a global ad network, and more.
|
Everything in one place New here? Start with the free chapter. | ||||
More from Metrics & Mayhem
|