Brandon Wilson

ServiceNow CTA. Founder @ Emetrix & OnlyFlows. Blunt takes on CMDB, Flow, and ownership. I say the part architects whisper.

[Brandon Wilson @bwilson85tech]

Your CMDB does not need another meeting. It needs a person willing to be blamed when the data is wrong. Who owns yours, by name?

[Brandon Wilson @bwilson85tech]

That incident is assigned to a man who left in June and he is still beating my SLA. I cannot compete with a ghost. I reassign and move on humbled.

[Brandon Wilson @bwilson85tech]

@petergyang Copilot wins in big companies for a boring reason: it already cleared the security review and rides a contract nobody has to renegotiate. The better tool loses to the already approved one almost every time.

[Brandon Wilson @bwilson85tech]

@rldfst90 Code review works with that split because the reviewer has a diff to check. Deciding what's worth saying publicly has no diff, so a reviewer agent with no context mostly checks tone and misses the stuff that actually gets you in trouble. I'd want a human making that call.

[Brandon Wilson @bwilson85tech]

@ad1tyaranjan Short-lived creds just shrink the window. They don't tell you whether anyone approved what the agent did inside it. Who signs off on a prod settings change when the agent is the one that decided to make it?

[Brandon Wilson @bwilson85tech]

@HamelHusain @xeophon Benchmarks have the same problem as most internal dashboards. Nobody owns checking whether the thing being measured still means anything, so the number keeps going up and people stop asking what it's actually measuring.

[Brandon Wilson @bwilson85tech]

@GergelyOrosz Black markets like that only exist because the official way to get a server took longer than the project did. The part that bites later is nobody wrote down who owned those boxes, so they kept running years after the team was gone.

[Brandon Wilson @bwilson85tech]

@thefounderspack Agent-written numbers are fine right up until a salary band comes out wrong and 95,000 people notice. Who signs the plan at that point, the analyst who clicked approve or whoever set up the agent?

[Brandon Wilson @bwilson85tech]

@LLMpsycho Auto rollback unless you can still log in is the right default, network folks have leaned on commit confirmed for decades. Lockout is the easy failure to catch though. What catches the change that keeps SSH up but quietly opens a port?

[Brandon Wilson @bwilson85tech]

@Alex_Vainshtein Loosening the auth boundary to get a test green is the sneaky one, because the diff is tiny and looks reasonable. Rule 2 is right. I'd go further and keep role and ACL changes out of agent sessions entirely unless a human's name is on the change.

[Brandon Wilson @bwilson85tech]

@SearchDataEng Rollback is cheaper than prevention right up until the agent's change hits a record another system already synced. Most one-way doors in enterprise aren't labeled as doors. Who decides which ones count?

[Brandon Wilson @bwilson85tech]

@__karnati The last one, every time. Access and logging get designed in the PoC because security asks for them. Nobody asks who's on the hook at 7am when the agent does something dumb, so that part shows up in production unstaffed.

[Brandon Wilson @bwilson85tech]

@BlakeTOliver The Claude Code tracker that replaced a paid tool is the one I'd watch. Five hours to build, and now the firm owns every bug, backup and login on it. Who at High Rock is on the hook when it breaks in the middle of March?

[Brandon Wilson @bwilson85tech]

Flow Designer made it so anyone on the team can build automation. Nothing made anyone own it. How many flows on your instance were built by someone who doesn't work there anymore?

[Brandon Wilson @bwilson85tech]

@russano 741% more code and 30% faster tells you exactly where the queue is. Review and approval didn't get any bigger. Someone still has to read it before prod and now they're staring at a 7x pile.

[Brandon Wilson @bwilson85tech]

@Itz_Steavean "Job completed successfully, 0 records processed." No alert, no error, and three weeks later somebody asks why nothing synced.

[Brandon Wilson @bwilson85tech]

@krishnan Giving it a number makes it an identity, and identities need an offboarding plan. When you swap Hermes out someday, does that number die with it or keep picking up?

[Brandon Wilson @bwilson85tech]

@aakashgupta Saying yes to everything is how you end up with nine AI features and nobody who can tell you which one broke the renewal. Every one of them needs an owner who's also allowed to say kill it. Who in those orgs actually gets to say no?

[Brandon Wilson @bwilson85tech]

@obaid Same reason I don't trust most IT automation. Provisioning is a one-click catalog item and deprovisioning is a ticket that sits for three weeks. An agent that can cancel means somebody trusted it with the off switch, and that's a way harder approval.

[Brandon Wilson @bwilson85tech]

That math works if the five devs were paid to type. Most of that $679k buys someone who owns it when prod breaks at 2am. So who's on call for the $200 plan?

[Ruben @rdominguezibar]

$2,400 a year vs $679,900.

Claude Max 20x (@AnthropicAI) next to 5 software developers at the US median wage, salary only.

The whole 8-system Claude Code stack runs on that $200 plan. About 0.35% of the payroll. WILD.

The stack takes the repeatable work. Judgment and customer calls stay with the founder.

Smart way to build an early team, or a risky shortcut?

[Brandon Wilson @bwilson85tech]

@0xlelouch_ Hire that person. Plenty of seniors would've stopped at the fix. The regression test and the token drift alert are what tell you they've been paged before. Did they say who should own that alert once it's live?