A developer who spent 30 years pairing Phish’s long-form jams with deep coding work finds agentic AI supervision has broken the flow state that defined his career.
Key Takeaways
The author’s flow state depended on sustained, single-context attention matching Phish’s extended jams; distributed systems and dissertation writing (200+ pages) were products of that ritual.
Agentic AI work is staccato: short responses, constant context-switching, parallel queues. The music that enabled focus now runs out of phase with the actual work rhythm.
This is not a complaint about competence loss. Useful output continues. The author’s grief is about the disappearance of subjective involvement and creativity that lived inside deep flow.
The shift from writing code to managing agents is framed as more leveraged engineering, but the author questions whether flow state and creative fulfillment can exist in a supervision-dominant workflow.
Hacker News Comment Review
Commenters split sharply on whether agentic coding is an upgrade or a loss. Those who disliked pure coding celebrate the shift; those who loved deep implementation mourn it, with several noting they now do personal or open source work to recover the feeling.
The “managing a junior who moves fast and breaks everything” framing for full-agent coding resonated widely. Commenters distinguished autocomplete-assisted coding (higher flow, still interactive) from full agentic (damage control mode).
A latency argument surfaced: LLM slowness forces multitasking, fragmenting attention the way batch-job eras did. Some commenters expect flow may partly return if agent speed improves, but tool-execution bottlenecks will persist.
Notable Comments
@kaiwn: “In the past we were forced to pour the concrete ourselves” – frames agent coding as finally separating engineering design from implementation labor.
@skybrian: Argues LLM latency is the structural cause of multitasking, not a personal failing; parallels mainframe batch-job and long-compile eras.
