Free live session
Your AI provider will fail. Design for it.
What happens to your architecture when the AI service is down, slow, or wrong.
Wednesday, September 30 · 10:00 am Pacific · 30 minutes · Live on LinkedIn
Save your seat
I'll send the link, one reminder, and the slides afterward.
Thirty minutes
Twenty of talk, ten of questions. No filler, no vendor pitch.
Six questions
A checklist you can take straight into your next design review.
Slides afterward
Sent to everyone who registers, whether you make it or not.
Someone wired up an API call and it worked.
The demo went beautifully. Product loved it, leadership loved it, and it shipped on a Thursday. Then it got transferred over to you to run.
Now there's a p99 nobody can defend, because the response time depends on how long an answer happens to be. There's a bill that moves with usage, which no other dependency in your stack does. And there's an on-call rotation that cannot tell the difference between the model being down and the model being confidently wrong.
None of that is a model problem. All of it is an architecture problem, and architecture problems are things you can deal with.
What you'll leave with
Six questions to ask about any AI dependency before it ships:
- Tier — have you given this dependency a criticality tier at all?
- Degradation — what does the feature do when the model is down?
- Latency — how long is the call allowed to take?
- Cost — what's the ceiling, and what happens when you hit it?
- Wrongness — how would you know the answer was wrong?
- Blast radius — what can it reach, and what can it change?
Two of them go deep, because they're the two your existing toolkit gives you nothing for. How would you know an answer was wrong? And what can the model actually reach?
Nothing about how a model works. All of it about what happens to your system when the thing you called doesn't behave the way you expect.
Wednesday, September 30
10:00 am Pacific, live on LinkedIn. Register and I'll send you the link, a reminder before it starts, and the slides afterward.
Who's running it
Lee Atchison writes about architecture, availability and risk at Software Architecture Insights. He's the author of Architecting for Scale and Overcoming IT Complexity, both from O'Reilly, and his courses have reached around 160,000 learners across Coursera, O'Reilly and LinkedIn Learning.
