I migrated https://t.co/gCWJNGBlOy from @nextjs Nextjs 15 to Tanstack Start @tan_stack in about 37 hours with a long Claude code session, ~ $728, 12 tickets, and 79 PRs created.
https://t.co/hi3YjfGeKj
I have been using Fable 5 for planning, mostly with Medium effort. When I have it build the plan, I ask it if it thinks Fable is needed to execute the plan, and what effort level is needed.
Because for most stuff you do, you don't need Fable to write the code, but planning is where Fable is absolutely brilliant.
In this example it suggests in a 7 part build to use Sonnet with an Effort level of Medium for Steps 1 - 4, and Opus, also at an Effort level of Medium, for Steps 5 and 6.
Try this, next time you put together a plan, here's the exact wording I use:
"I want to [ write what you'd normally write when building a plan ], let me know if I need low, medium, or high effort to do the actual implementation, and if I need Fable, or if I can use Sonnet or Opus, or both."
Stop writing so much code with Fable 5, you don't need to, but you should build incredibly strong plans with it, and use it to code when Fable itself thinks you need to.
(note: always ignore those estimates in days, these are usually way off, it's always much faster)
1/ Brainstorm with Claude Cowork to get more ideas and product vision.
2/ Ask it to write them all down in plans/.
3/ Claude Code work overnight on the /goal and getting 50+ PRs merged next morning.
Claude Sonnet 5 is now on DeepSWE
It scores below Opus 4.8, costs twice as much, and is even more expensive than Fable 5
Probably Anthropicโs worst release yet
Paper Shaders: now fully open source and FREE
Yes, you can make a: plugin, tool, template, site, app, library. Resell it. Make a Figma plugin. It's already code. Go for it!
We're pushing shaders forward. New Paper Shaders on the way.
I am not ready for real user yet, everything broken and not clear the core vision, but the product name is good by itself. I haven't done anything for marketing.
Introducing TabFM, a foundation model designed specifically for tabular data classification & regression. This approach allows generation of high-quality predictions on previously unseen tables in a single forward pass.
Learn more and try out the model โhttps://t.co/OTbVQ8oUQs