20+ years writing code, now building AI agents beyond the demos. Engineering leader sharing real-world lessons on voice AI, agent workflows, security and trust.
I see posts from friends regarding dealing with bad Voice AI experiences. Sometimes you just gotta bring your own Voice AI to the table! @bot can now make calls for your with your phone number!
@Cristina_AC_X Agreed. Being a leader does not require a title. I've had coworkers in my career who were more influencial then those with the big titles. Humble, smart, emotionally regulated.
@shayneparlo Marketing has ramped up for sure around new products. I typically just go to the vendors trust center to see the underlying providers. Either way, Livekit imo has solid features in this space for everyone to utilize, vendor agnostic so to speak.
@emollick I've been paid to write code since 2005 and I've watched whole stacks disappear. The difference is those took years and you could see them coming. This one isn't giving people that kind of runway imo.
@theo I'll read less of it, sure. But I'm still not letting AI approve AI code. Pre and post hooks, build checks in the pipeline, some boring deterministic gate at the end, if that works consistently then that will probably be what causes me stop reading code altogether.
@RoundtableSpace 100% agree. The teams I've seen do this well run their own evals on every new model before switching. Prompts don't always carry over between models either, so your own tasks show you what to adjust way faster than a leaderboard.
@swyx Congrats on the milestone! Not surprised, coding agent practices wins. Everyone's seen the demos, what people really want is how to run these day to day without making a mess.
This is huge @SierraPlatform, This is the piece personal AI has been missing. I've had my @bot agent handle phone trees and cancel orders for me this week, and honestly the only reason I trust it is I can see what it's allowed to do. Identity and permissions first, then the fun stuff!
@parthnradia Unintentional or intentional A2A is going to explode over the next 6 months imo. When people understand that their personal AI, Grok, Muse, Dot or otherwise can handle those A2A situations for them.
Used it several times this week, and it was able to navigate the voice AI agent phone tree of Spectrum and today when there was no cancel option in the @SamsClub app for a custom bakery order my wife sent to the wrong location I had my @bot agent call the bakery and cancel while my wife went back into the app to order from the correct location. It's phenomenal!!
The press-zero habit is earned. Most of those phone trees wasted people's time. That's why the move isn't yelling "human" louder, it's bringing your own agent to the call.
This week one of my @bot agents handled several calls for me with companies that have AI voice agents on their main number. Agent to agent, task done, I never touched the keypad. Trust is the whole product.
"You are the long-term maintainer" is a great framing, but prompts don't always carry over between models. The most successful teams I've seen run evals on every new model before switching, so they know exactly what to adjust to keep getting the results the last model gave them. Swapping models without evals is just vibes.
I agree, as someone who took over half a dozen whiteboard interviews in my career as a software engineer. It was a requirement but it really didn't show you the curious, the persistent, the hungry to learn SWE's. Just those that knew the test material.
The coding interview is dead, at least for my team at Anthropic. I've seen enough to know that it no longer provides signal. I removed it from loops this month.
I have a spicy, optimistic take - I think there's an opportunity for junior engineers: https://t.co/PVvpQCbQtK