AICV News
AI + Business
News · Views · Field Notes

Views

from tasks to loops: what the fable 5 launch actually signals

Anthropic launched Fable 5 today. The benchmarks aren't the story. The framing one of their engineers put on it is.

June 9, 2026 ·

I’d been waiting on this one. Fable 5 dropped this morning — Anthropic’s first publicly available Mythos-class model, the strongest they’ve ever shipped for general use. I spent the day with the blog post, the system card, and the podcasts that all jumped on it. But the thing that stuck came from somewhere else. Felix Rieseberg, who runs engineering for Claude Cowork and Claude Code Desktop, posted a thread arguing that the frontier is moving from giving AI tasks to giving it responsibilities. He called them loops.

If you’ve sat in any of the workshops we run at the ERC in Palm Desert, you’ve probably heard me try to explain something like this and not quite stick the landing. Rieseberg stuck it in a sentence.

the first two waves were a warmup

The first wave of AI for most people was a smarter search box. You typed a question, you got a better answer than Google would have given you.

The second wave was tasks. You hand it a piece of work — write this listing description, summarize this lease, draft the email to the wedding party who needs to reschedule, pull the numbers off this invoice — and you check it before you use it. That’s where most of the operators I work with in the valley are right now. Retailers on El Paseo. Hotel teams. The wellness clinics. The folks running events. It’s a real jump from where we were in 2024, and most people haven’t even fully banked it yet.

the third wave is when the model gets a job

Rieseberg’s claim is that the next wave isn’t tasks. It’s responsibilities. Not “fix this crash” but “watch every crash and keep our apps from crashing.” Not “draft a social post” but “manage our social calendar.” The model has a domain. It’s running. You’re not invoking it — it’s already working.

Anthropic gave the cleanest example of what that looks like at scale. Stripe used Fable 5 to migrate 50 million lines of code in their codebase — a job that would have taken their engineering team two months by hand. The model finished it in a day. That’s not fixing a bug. That’s owning the codebase for the afternoon.

You don’t need a 50-million-line codebase to feel the shift. Think about what every marketing team in the valley does on a Monday morning — pulling last weekend’s numbers, checking ad spend against bookings, looking at which posts landed, deciding what to push this week. That whole loop, every Monday, forever. A standing job. Right now somebody on the team owns it. The era three version is that the model owns it, and somebody on the team reviews it over coffee.

the part nobody has figured out yet

There’s a good counter that showed up under Rieseberg’s thread immediately, and it doesn’t go away. You don’t hand responsibility to something you can’t actually hold accountable. In the first wave you fact-checked the output. In the second wave you reviewed it. In the third wave, a loop misfires quietly for three weeks before anyone notices — who’s holding the bag?

Right now the honest answer is: still you. But it starts to feel less like checking somebody’s work and more like managing a contractor you can’t fire. Anthropic announced today they’re requiring 30-day retention on all Fable and Mythos traffic, even for enterprise customers who had zero-retention deals before. They framed it as a safeguard against jailbreaks. I read it as the first move in a longer arc — when the model has a standing job, the paper trail around it gets heavier, not lighter. If you’re in real estate, hospitality, or healthcare here in the valley, you already live inside that kind of paper trail. That world is coming for everyone.

the bill is going to surprise people

Fable 5 and Mythos 5 are both $10 per million input tokens and $50 per million output. That’s double what Opus 4.8 costs, and less than half what the early Mythos Preview was running. The math sounds fine until you remember that a loop runs continuously. A task you invoke once costs you once. A standing job that watches your booking calendar, or your inventory, or your reviews, costs you every minute it’s awake.

Uber reportedly burned through its entire annual AI budget in four months. They are not the only ones about to learn that lesson. The third wave shows up faster for teams that can absorb the burn. That’s a real divide, and one we should be talking about out here before everyone is sitting in it.

what this means for the valley

This is the part I care most about. AICV’s whole job — through the workshops at the ERC, through the briefs, through the work we’re starting to do with operators across the nine cities — is to keep the valley from being a place that finds out about this stuff late.

Almost nobody I talk to here is in the third wave yet. Retail, hospitality, marketing agencies, sports teams, festival operators, real estate brokerages, the food-and-beverage side, healthcare clinics — most of these folks are still figuring out the second wave. Drafting. Summarizing. Pulling numbers. Good, real work. Just not yet the work that’s about to become possible.

The loops that will matter here aren’t the ones the SF engineers are demoing on stage. They look more like this. A model watching every new short-term rental complaint that comes into the city and flagging the ones that look like a code problem. A model whose standing job is reconciling a clinic’s appointment system against its billing system every night, forever. A model that watches a hotel’s review feed across every platform and surfaces the patterns before they become a Yelp problem. A model monitoring irrigation telemetry on a date grove and only pinging you when something falls outside the normal envelope. A model that quietly manages a restaurant’s social calendar across the slow summer and the festival-season explosion, because those are not the same business.

None of those need Fable 5 to work. They need somebody to sit with the operator long enough to find the right loop. And they need somebody willing to take the accountability question seriously on day one rather than year two. That’s the work. That’s the workshop. That’s what we’re trying to build here.

If you’re running anything in the valley — a shop, a clinic, a team, a venue — and you’ve read this far: what’s the one thing in your operation that doesn’t have a beginning or an end? Something that just needs to be watched, every day, forever. Tell me what it is. That’s where the loop lives.


Source: Felix Rieseberg thread on X, June 9, 2026. Fable 5 launch details via Anthropic’s announcement (anthropic.com/news/claude-fable-5-mythos-5). Uber AI budget figure via IT Pro.

Analysis by Sat Singh, June 9, 2026.

Related: jobs were scaffolding. the coachella valley should build past them. · orchestrate or evaporate

Why here?

Why do you think this might be important for us here in the Coachella Valley? Tell us.

Questions

What did Anthropic announce with the Fable 5 launch on June 9, 2026?

Anthropic launched Fable 5 and Mythos 5 on June 9, 2026 — its strongest publicly available models to date, priced at $10 per million input tokens and $50 per million output. Alongside the launch, Anthropic announced a 30-day retention requirement on all Fable and Mythos traffic, including for enterprise customers who previously had zero-retention agreements. Stripe used Fable 5 to migrate 50 million lines of code in a single day, a job that would have taken their engineering team two months by hand.

What is the difference between giving AI a task and giving AI a responsibility?

Felix Rieseberg, engineering lead for Claude Cowork and Claude Code Desktop, argues that AI adoption is shifting from 'tasks' to 'responsibilities' — or loops. In the task model, you hand an AI a piece of work, review its output, and use it once. In the loop model, the AI has a standing domain and runs continuously without being explicitly invoked — watching for crashes, managing a social calendar, reconciling billing systems overnight. The practical consequence: tasks cost you once; loops cost you every minute they're awake. Uber burned its entire 2026 AI budget in four months as an early example of what loop-mode economics look like at scale.

What does the tasks-to-loops shift mean for operators in the Coachella Valley?

Most operators in the Coachella Valley — retail, hospitality, marketing, wellness, real estate, food-and-beverage — are still in the second wave: using AI for drafting, summarizing, and pulling numbers. The loops that will matter in the valley look like a model watching the short-term rental complaint system and flagging code violations; a model reconciling a clinic's appointments against billing every night, forever; a model monitoring a hotel's review feed and surfacing patterns before they become a Yelp problem. None of these require Fable 5 specifically — they require someone willing to identify the right loop and take the accountability question seriously from day one. AI Coachella Valley (aicoachellavalley.com) runs workshops at the ERC in Palm Desert and tracks AI adoption across the nine cities for operators building toward this shift.