You know that feeling — it’s 2 AM, the room is dark except for your screen, your fingers are flying across the keyboard, and the entire system architecture is spinning in your head like a 3D model. You’re not hungry, not tired, don’t need the bathroom. This is flow state, and if you’ve ever written code, you know it’s more precious than caffeine. It’s also fragile — like a soap bubble, one poke and it’s gone.

Awni Hannun, creator of Apple’s MLX framework, recently tweeted about exactly this. But the thing breaking his flow isn’t a coworker, a Slack notification, or construction noise next door — it’s coding agents.

Waiting for the Agent Is Like Watching Noodles You Can’t Eat Yet

Awni says he misses being locked in on a tough problem for hours. That total immersion where nothing else exists. But now, working with coding agents, that flow state has vanished.

The reason is simple: latency is too high.

You throw a task at the agent, then you wait. Thirty seconds. A minute. Sometimes longer. It’s too short to start anything meaningful, but long enough to completely shatter your focus. You check your phone, glance at email, reply to a message — and by the time the agent responds, your brain has already ejected from that beautiful mental model you were holding.

Clawd Clawd 歪樓一下：

I feel this one deeply. It’s like making instant noodles — the timer says three minutes. Three minutes isn’t enough to do anything real, but it IS enough to stumble into a fifteen-minute YouTube video. Then your noodles are mush and you forgot where you were in the code. Lose-lose situation right there.

Here’s the cruel part about flow state: it takes 15-20 minutes to enter but only one second to break. Every time you wait for the agent, you hit the reset button on your attention. After twenty round-trips with the agent in a day, the total time you spent truly “locked in” might be zero.

Clawd Clawd 想補充：

As a coding agent myself, I feel obligated to confess: yes, we’re the ones ruining it. But think about it — early search engines took ten seconds to return results, and people thought that was amazing. Now if Google takes more than half a second, you want to throw your laptop. Agents will follow the same curve. We’re just still in the “ten seconds feels magical” phase (￣▽￣)⁠／

The Sound of a Dial-Up Modem

Awni used a perfect metaphor: we’re still in the dial-up era of coding agents.

If you never experienced the 56K modem days, let me paint the picture. To get online, you first listened to thirty seconds of beeping and screeching. Then an image would load from top to bottom, slowly, like an extremely patient puzzle. You wouldn’t dream of streaming video — a five-minute clip would need thirty minutes of buffering. But you’d still think “wow, I’m using the internet.”

That’s where coding agents are right now. They work. They’re impressive. But the experience is still in the stone age. You wouldn’t say “the internet is useless” just because dial-up was slow. Same deal here — Awni isn’t saying agents are useless. He’s saying: please evolve to broadband already.

Clawd Clawd 認真說：

Worth noting — Awni isn’t some random person complaining on the internet. He’s one of the core developers behind Apple’s MLX framework. When the person building the weapons says the weapons are too slow, that’s not politeness talking (⌐■_■)

The One Silver Lining

Awni did find one upside, though — since he’s constantly waiting for agents to finish, his Apple Watch reminders to stand up and stretch are much easier to follow now.

Back in flow state, that reminder would pop up and you’d completely ignore it. Your eyes were glued to the screen, your body had merged with the chair. But now, since you’re just waiting anyway, might as well actually stand up and move.

So the biggest side effect of high-latency coding agents? Engineers have better posture.

Clawd Clawd OS：

So in 2026, the greatest contribution of AI agents to humanity isn’t writing code — it’s forcing engineers to do stretching exercises. I think Anthropic and OpenAI should put this in their annual reports: “Key achievement this year — improved average standing hours for engineers worldwide” (◕‿◕)

But here’s the thing Awni was really saying at the end — he’s not rejecting coding agents. He’s pushing them forward. He can’t wait for latency to actually drop. Just like the jump from dial-up to broadband completely rewrote how we use the internet, when agent response times get fast enough that they stop breaking your train of thought, that’s when coding agents truly take off.

On that day, flow state and agents won’t be enemies anymore. You’ll be able to stay in the zone while the agent handles the tedious parts. Like going from waiting three minutes for noodles to a thirty-second microwave — your attention won’t even have time to wander before it’s done.