Claude Fable can run autonomously for days.
This is the single highest-leverage prompt I use to stay on top of it:
"Spin up a persistent HTML page. As you work, append clear, timestamped updates with screenshots/media so I can follow along."
Literally a 10x better experience.
Many users called the persistent Claude agent tracking prompt with its HTML dashboard a game changer for workflows and parallel management, while some objected to the high token costs and frequent model downgrades.
This works anywhere that gives you a persistent URL... I have it use Railway, but you can customize the prompt to have Claude use whatever platform you like best!
Claude Fable can run autonomously for days.
This is the single highest-leverage prompt I use to stay on top of it:
"Spin up a persistent HTML page. As you work, append clear, timestamped updates with screenshots/media so I can follow along."
Literally a 10x better experience.
@BlasikRandy @Trace_Cohen I haven’t had this issue at all, totally depends on the work you’re doing with it (I’ve mostly been using it for 3D/game design)
@mattshumer_ This is genius.
HTML for ridiculously clear presentation of the work done & timestamped with updates to see the flow & sequence.
@WinterArc2125 It’s genuinely a huge unlock. Allows me to manage many more agents in parallel, as I can re-orient around the right context much more quickly.
@mattshumer_ @mattshumer_ that's pretty genius tbh. gonna try it with my own projects, sounds like a game changer for keeping track of progress without getting lost in the weeds. thanks for sharing!
@mattshumer_ @Trace_Cohen Love you bud but B.S. I can't get fable to run for more than 5 minutes before it switches to 4.8 🧑💻🧐
@mattshumer_ Persistent context trail that you can actually read back is underrated. How far have you pushed it before it starts losing coherence on longer runs?
Execution without persistent telemetry is just fast-paced automated technical debt. By forcing the agent to build its own visual dashboard, you aren't just managing the work—you are building a continuous cybernetic feedback loop. The next wave of enterprise architecture isn't about giving agents better tools; it’s about giving humans better ways to audit machine speed.
@mattshumer_ But can your bank account run Fable for days? Amirite??
@mattshumer_ Can we build a consultant group Matt
@mattshumer_ I built persistent hook-informed timeline capabilities into atrium for this very reason, but this is inspiring 💡
@mattshumer_
@mattshumer_ For real.
Thanks for the workflow!
Fable seems to be very good on task involving vision/visual stuff (HTML / Animations / Remotion / Manim).
@mattshumer_ why would you not have it make it a wiki with a roadmap, epics, tasks, and whatever artifacts, links, and open items for the human that it updates on an html page. then you can navigate the whole project instead of trying to fight a blob. just make it update it after each task.
@mattshumer_ The single highest-leverage prompt is... making it print HTML nicely. That tracks.
@mattshumer_ 10x experience= 100x expensive
@mattshumer_ have you tried it with opus ?
@mattshumer_ A persistent log page changes everything. Instead of scrolling through chat history, you get a dashboard. That is a simple idea that works really well.
@mattshumer_ @Trace_Cohen I'm not doing anything crazy just wanted it to go through my codebase and close a few issues on github (and it simply switches to 4.8 for "security reasons".
@mattshumer_ Now this might be my favorite workflow idea, out of everything in the past year. Amazing find
Claude Fable can run autonomously for days.
This is the single highest-leverage prompt I use to stay on top of it:
"Spin up a persistent HTML page. As you work, append clear, timestamped updates with screenshots/media so I can follow along."
Literally a 10x better experience.
