4 Comments
User's avatar
Alex Ponomarev's avatar

Great read, Alex! This also creates an interesting leadership problem.

Making AI reliable is mostly invisible work, and when everything runs smoothly, it looks effortless to everyone who didn't build it. SREs spend their time adding guardrails and fallbacks around stochastic models, but nobody notices unless something goes wrong.

Alex Ewerlöf's avatar

Exactly. And I'd argue the masses still don't know the SRE's toolbox. And the SREs I know are too chill to shout about it. Regardless it bugs me that many of these old patterns are being "rediscovered" and sold as new ground breaking sokutions (while conveniently ignoring the trade-offs, which tells me some expensive lessons are due to be learned).

Alex Ponomarev's avatar

Maybe the rebrand might be doing some good too? Loads of people building with AI would never touch an SRE book, but they’ll happily grab an “agentic loop.” So it ends up with people who needed it and had no idea it existed. Doesn’t fix your trade-offs thing though. But I guess this just means more people get to learn the hard way this time around.

Aleksey Shevchuk's avatar

Totally agree. AI taming and all the buzzwords recently created have elder siblings in old-fashioned engineering we've been doing for years. Rejecting the hype and reflecting on which known approach to apply enables mindful advancement.