⚡️⚡️⚡️Building LineAI: a runtime where a manager agent builds, tests, and runs your speech-to-speech voice agents. Technical founder (1 exit, 1 to 3M MAU)⚡️⚡️⚡️

San Francisco
Joined October 2008
Replying to @not_ellington
In reality these models cannot implement a simple endpoint without adding unneeded backwards compatibility, migration logic that has nothing to migrate, plus zero concurrency so it runs as slow as possible. They are not taking over anything
2
1
388
Depends on what you’re working on. They are impressive if you’re building something new. Less so in a 10 year old codebase with things that break. And they are so easy to trip up… one out of date readme and it is off solving a non-existent problem. Fragile and flaky
10
Why are devs the only ones smart enough to get AI to do their jobs for them and still be every bit as valuable as before?
1
4
38
Claude has gotten so smart what used to take hours now takes days 👏🏽
49
Hahahaha, Anthropic must be trolling its users. X: Opus is unusable use Fable 5 Fable 5:
37
The comments 😂 It’s a Claude therapy session. - And that’s not the load bearing part”
Wife: "You picked up the kids from school, right?" Me: "Three findings, and one of them is more serious than the question you asked."
42
Is this your king?!?! Even Fable 5 cannot run safely, unmonitored.
18
Don't connect Gmail to your agents! It gets disconnected w/o warning, and your email has too much in it to risk getting hacked. Every LineAI voice agent gets its own native email that just works. Its the agent's phone number, like this: [email protected]. Here's how:
70
Are people really connecting Gmail to their agents? Sounds like a hackers dream
1
42
Taking a breather after this epic one-year run building LineAI from scratch. Hit $100K ARR pre-launch. Being 40+ I knew I would have to build my own future so I started and went all-in
1
3
78
Qwen3.8-27B-4bit and Qwen3.8-27B-8bit on oMLX with DFlash 2. M3 Max. I was able to hit 32 tok/s on the 4bit only once then sub-20 since. Good enough to just let it run while I sleep. So jealous of the lucky few with the 5090
1
78
I run claude-sdk in the cloud and use it to create and manage all of my 100+ voice agents. In this video it creates an agent to answer after-hours calls for commercial real estate. The manager runs on Claude-sdk with a custom runtime that lets it do everything that I can do when I build voice agents. And it can launch dialer agents to call, verify each requirement, and iterate.
2
3
170
This lets you see exactly how your realtime/speech-to-speech agent's context is doing turn-by-turn
2
53
This voice agent will not give away the owners schedule, or transfer unsolicited spam callers, but it will patiently explain the menu, email reservations, upsell private dinning for groups, and tell anyone calling about a yelp optimization review to fuck off. All verifiable before it starts taking real calls. I built this platform to make it easy for almost anyone to launch agents that are ready for hard calls in the real world
1
44
This is super easy - A commercial real estate STS voice agent with local phone number in a minute
19
I built LineAI to help teams launch speech-to-speech voice agents (the only ones that actually hear) that work consistently in the real world. The difference between a demo and a real world agent is our harness and pro tooling designed for audio not text. getline.ai/try
1
74
Opus 5, Sonnet 5, Fable 5 are destroying Anthropic
Why are Anthropic models seeing a decline in usage on OpenRouter? What could be the reason?
58
Great sounding idea that works on X for engagement. but it won’t fix Opus 5.
Experimenting with a rare addition to my global CLAUDE.md: "Always talk in ASD-STE100 Simplified Technical English. Always read CONTEXT.md files, and use their ubiquitous language."
1
28
Opus used to feel relaxing. Opus 5 is like riding a bull
Am I dumb or are agents getting harder to understand? I feel like my mind is no longer able to keep up with the massive context they work with and the summary output feels like reading another language
56
One more 15-day streak and we launch
2
2
39