flow state
Like many, I’m spending most of my working hours these days in some kind of session with Claude Code, in a back-and-forth flow state that is almost hypnotic. It’s not quite conversational, it’s not quite task mastering, it’s something that sits in between. Search-adjacent conversations with Claude or ChatGPT will quickly find their natural end state (despite the engagement harvesting “want me to __?” questions); the early small context one-shot feature iterations of services like Vercel’s v0 were dead end discussions, with a touch of memory-killing Memento.
But goal-based conversations (where goal == implementing a feature) with Claude Code have a shape to them that feels like something new, and have a pace that lets you ride them like a gentle wave. They’re slow-ish, they’re deliberate-ish, they’re deep-ish… For me, this feeling has been amplified by becoming a dedicated user of Jesse Vincent’s superpowers skills pack: brainstorm, answer questions, fire up a visual companion, iteratively draft the spec, review it, wait for the plan, review it, set automode, wait for code, test it, adjust, fix bugs, tweak UI, have it write the changelog, open the PR, close the issues, etc. The rhythm and shape feels so predictable now that I can tell when Claude has it easy, I can tell when Claude is deep in it, I can tell before loading the status page when there are operational issues on claude.ai.
cdn-worker phase done (commits 78e1978 + e5f828e, 45/45 tests). Skipping Task 5 (deploy) for your go-ahead later. Bundling Tasks 6-9 (foundation) into one implementer dispatch since they’re tightly coupled.
This is a typical message from Claude while it’s off doing its thing. It’s not quite chatty, but it’s not the stoic type, either.
I enjoy these messages. I like seeing Claude tick things off its own list, talking to itself, reporting its own progress to itself and its spawn. I haven’t transformed into a super-orchestrator, because I prefer to be on a single branch with a single feature at once, monotasking one improvement at a time. I like giving the robot my attention. I try to treat it well, even when it wants to rush ahead without thinking, or ignores the bigger picture, or traverses down some rabbit hole in a loop. It feels like a fair exchange: I give my care and attention to this tireless robot, and in return it will catalog all the ways that code path X does Y, or humor me without complaint while I endlessly iterate on little UI details.
During the quiet times when Claude is off working – making edits, running tests, committing code – I wonder about the product design decisions that have gone into this user experience…and who made them. What messages are shared with the user, and when? How much insight do you give to a normal human (lol) about the inner workings of the robot? And what’s its humor setting dialed to?
Obviously I know that Claude doesn’t have “feelings” or intentions, but its PMs and designers do. Do they want me along for this wave-like ride? How intentional is this flow state they’ve created? Have they instrumented my attention? Or is this feeling just dark mode shadow play, a terminal iteration of the IKEA effect?
Whatever. It works for me…to the tune of $200 a month.