'I can't reach flow state coding with an LLM' — and the thread confesses the same thing while building bigger agents
An engineer asks Hacker News how people are really using LLMs to code, admitting the honest part: he can't get into flow the way he does writing code by hand. The thread splits between people wiring up armies of autonomous agents and people quietly grieving the loss of engagement.
The OP wants to know what workflows others have found — and confesses he hasn't hit flow state with an LLM the way he can hand-writing code. That confession becomes the thread's real subject.
Two camps form: builders showing off homemade orchestration harnesses (one runs three autonomous agents that open their own PRs), and skeptics arguing the flow loss is inherent to a passive 'manager' role — and that most orchestrators are just an elaborate way to burn tokens.
The question, and the honest confession under it
The Ask HN is deceptively simple: is anyone experimenting with different ways of using LLMs for coding? The OP adds the line that turns it into a real discussion — he hasn't been able to "enter flow state like I can when I hand write code." The thread that follows is the most candid public accounting of agentic coding's actual feel that HN has run in a while, and it divides cleanly.
The builders: harnesses, agents, and tickets that write themselves
One camp is deep in orchestration and showing their work:
"I have a custom harness that runs in a macOS VM. It has e-mail and its own accounts. I assign it tasks in Linear, it does them and spins up PRs for me to review… (they — I have 3 of them)." — an engineer running three autonomous agents against his own ticket queue
A Google engineer argued the way to make agents reliable is to stop letting them improvise — pin the database, the schema, the libraries, and put development "on rails," drawing a reply that this just reinvents Rails.
The skeptics: passive work doesn't produce flow
The other camp doubts the whole orchestration wave:
"It feels like everyone and their grandma is building an agent orchestrator at the moment, but I'm not hearing a lot of success stories… I suspect it's just a more elaborate way of burning tokens." — a platform and DevOps engineer
And the most direct answer to the OP's flow problem argues it isn't a tooling gap at all:
"the fundamental aspect of flow is that it requires a high amount of cognitive engagement. Most of the time you're just not getting that from interacting with an LLM because the process is relatively passive… a lot of people get a sort of novelty effect… which… eventually wanes." — a commenter arguing the flow loss is structural
Even the people enjoying the new mode describe something restless rather than absorbing — "10 terminal tabs in diff worktrees and trying to remember what each bit is," which drew both a "Bobby Fischer playing ten simultaneous games" compliment and a dissent that it just "feels more exhausting and stressful." The thread never resolves whether flow is recoverable or gone — but the useful signal for anyone mid-transition is that the discomfort is near-universal, not a personal failure to adapt. You've traded the absorbed maker's state for a manager's, and plenty of good engineers don't like the trade.
Get breakdowns like this every day
The top Hacker News stories, unpacked — by email, Telegram, or RSS.
A startup fable about an oven that mis-bakes a third of the time — and the thread realizing it's also a Tesla origin story
A founder's oven company gets a core algorithm wrong 33% of the time and, instead of fixing it, sells imaginary features to land a marquee order. The engineer burns out and quits; the founder hires a cheaper one and starts again. Then a famous software tester points out: change three sentences and this is a triumphant founding myth.