CEO @coworkerapp previously @uber
San Francisco, CA
Joined January 2011
- Tweets278
- Following882
- Followers238
- Likes337
Pinned Tweet
Today, we dropped the price of enterprise AI by 80%.
Same frontier AI. Same chat, cowork, and code experience.
Just 5x more tokens for the same spend.
Here’s how and why. 👇 Also, we made a video. Please enjoy.
9x cheaper, 64% faster, preferred on quality 84.5% of the time. No change to your existing workflows.
Coworker lets you chat, cowork and code with your organisational context across any model, open or closed. What a time to be alive.
It is clear open source models and harnesses are having a moment. There's a few factors at work
1/ It is now obvious that you can catch up to near-SOTA performance and do so with a clear training lineage. See:@thinkymachines Inkling launch today.
2/ There are several well-funded, talented teams building open weight models now in the US and abroad. Along with the explosing of other near SOTA models (Grok/Cursor, Muse Spark), it is clear we are going to have a diverse ecosystem of models atleast on coding and agentic use.
3/ Organizations are increasingly looking for control over how their data is used and are willing to trade off some access to frontier level tokens for this control. Organizations and countries are increasingly nervous about the frontier labs potentially competing with them down the road and don't want their data to enable a future competitor.
4/ Open source is a slider: you could bring your own open harness, your evals, your business context and are free to pick and choose your model of choice.
5/ Companies have now actively shifted from "how do we get our people to use tokens" to being uncomfortable with their token cost ballooning without a clear line to revenue.
6/ Geo-politically, countries will be weighing open weight models as a way to get frontier-level tokens inside controlled environments that may not be otherwise possible.
All of this leads to more choice for all of us !
1
5
229
Given this is a 2.8T model, my guess is capabilities land somewhere around Fable (and well ahead of Opus 4.8). That is insane for an open model.
1
127
Tea is being thrown into the harbor
🚨 YOU'RE OVERPAYING FOR AI BY 5X. AND ALMOST NOBODY HAS NOTICED.
every task you run goes to the most expensive model, even when an open model one nails it for a fraction of the price (yes, the one china dropped during the Fable 5 ban benchmarks near Opus 4.8).
You can fix it and save thousands with this new tool:
→ it routes each task to the right model, open or closed, automatically
→ same frontier quality, ~5x less in tokens
→ ask for a deck, dashboard, doc or app, get better results with less.
I ran it on real client work and they couldn't tell the difference. except in the bill lol.
stop paying frontier prices for intern tasks.
We kept hearing the same problem from Codex and Cowork users:
We like building stuff, but all that context blows a hole in token budgets.
Open Artifacts routes the work to the right model and context, for 4-5x less cost.
The outputs are genuinely beautiful.
Monty Python with a consolation prize youtube.com/watch?v=PrvXoin9…
1
41
What @coinbase did is impressive but 99% of companies can't do this.
@coworkerapp let's any company do this out-of-the-box
How to keep AI spend flat while token usage grows exponentially: Not with friction and spend alerts. With better defaults, routing, and caching.
Better Defaults (not Usage Caps) – Engineers can choose any model they want, but defaults matter. We’re experimenting with defaulting to open weight models like GLM 5.2 and Kimi 2.7 through our LLM gateway, while still encouraging engineers to choose the right model for the task. 91% of our employees were never hitting their usage caps, so instead of lowering caps and driving up alerts, we're moving to cheaper defaults. Note that code reviews use a diversity of models, so they can check each other's work.
Better Routing – In our custom harnesses, we preprocess prompts and route to the best model for the job, considering cache hits and model pricing. For instance, you may want a frontier model for planning, but not for execution where they can be overkill. Ultimately, humans shouldn't be choosing models - AI can automate this task.
Better Caching – Cache misses are the easiest way to drive your cost up. All of our requests are cache aware, so we’re reusing a warm cache wherever possible. For example, our cache hit rate went from 5% → 60% in LibreChat once properly implemented.
Keep Context Lean – Start fresh sessions when switching tasks. Scope file context narrowly. Disconnect unused tools. Don't just compact. The goal isn't fewer tokens used, it's fewer tokens wasted.
Better Visibility – Our engineers can use as many tokens as they want, from whatever model they want, but we’ve made usage visible – and the more you spend on AI, the more impact we expect.
The goal isn't to suppress usage. It's to build the infrastructure that makes exponential growth sustainable.
Putting this into practice has cut our AI spend nearly in half, while our token usage continues to grow.
1
1
127
Most of the convo on open weight models has centred on a handful of AI-native companies post-training models on their data (Fin, Cursor etc), and this being the path others will follow.
The vast majority of companies won’t do this.
Yet they’ll still have strong incentives to adopt open models alongside anthropic/openai (since open models are 10x cheaper at or near frontier performance and also to avoid lock-in).
This is where we’re going to see an explosion of companies buying context and routing layers: platforms that let them generate portable context across their apps, agents, and employees’ AI use and route it across open, closed and even on device models.
And the Fable convo shows that this isn’t just about maximising token budgets or providing an easy switchboard across the latest models.
What’s even more important to companies is their data and IP sovereignty: ensuring that they own the incredibly valuable shared context layer that captures their company’s workflows, IP and how employees work with and alongside AI and agents.
Companies just can’t risk a single AI lab owning that piece of infrastructure and being blocked or price gouged on something that’s becoming existential to their IP and how they do business.
1
4
69
The rebel alliance wins if I recall correctly
For the first time in ~3 years, it feels like the AI table has been flipped over. Yes, the labs and hyperscalers will have the highest chance of resetting it before everyone else, given their vast capital, frontier model, and compute advantages.
But there is now a window for a new ecosystem to emerge. A "rebel alliance," which is what we're calling it at @USV.
We are excited by open weight models, distributed compute, human-aligned agents, routing, open source harnesses, the orchestration layer, and a lot more.
Basically, anything that gives people and enterprises powerful intelligence while maintaining tight incentive alignment, we're into it.
2
71
Worth reading
If you understand my post this means I am bullish the inference providers like @AskVenice
Disc. Long VVV
1
52
Enter companies like @coworkerapp that pair every task with the right context and model to do the job at frontier quality for the lowest cost.
Sam Altman said AI budgeting has recently become a "huge issue" for some companies, something that "never came up" earlier this year. bit.ly/4uxIGnv
2
204
When you ask AI to do something, you’re contracting out work. Today there are 1-2 contractors.
Open models launching at frontier quality, eg kimi k2.6, is going to make this a giant marketplace problem.
As token budgets take on a larger part of operating expenses over time, model routing is the inevitable conclusion. This is also one of the biggest areas of differentiation for the applied AI layer over time.
By understanding the different work patterns in your domain, and having strong evals for that domain, you’ll be able to cost/performance optimize effectively.
We’re still likely at the point where most use-cases will need frontier performance for the foreseeable future; but soon you will be able to peel off individual use-cases and send them to lower cost models once the quality is sufficient for the task.
Enterprises individually trying to figure this out themselves at scale will likely not be possible, so the products that can intelligently route these workflows to the right tier of model will be in a strong position to aggregate more demand.
2
6
478
@BrendanFoody we should chat
"Right now we're spending more on tokens for our internal agents than we are on employee head count," Mercor's CEO Brendan Foody said. bit.ly/4uObeKg
69
👀 why limit budget when you can get 5x frontier tokens for the same spend by dynamically routing across open and closed models. @coworkerapp
Uber says it has limited all employees to $1,500 in monthly token spending per AI coding tool "to responsibly encourage agentic AI adoption" (@natlungfy / Bloomberg)
(Visit Techmeme dot com for the link and full context!)
2
116
this is where products like @coworkerapp that dynamically route across open and closed models while maintaining enterprise security and context are huge. 80% token cost savings for same quality outputs
*UBER SETS $1,500 MONTHLY CAP ON SOME AI CODING TOOLS FOR STAFF
$UBER officially reeling in the Claude budget after blowing their AI budget earlier this year.
Undoubtedly more companies to follow
1
126
enter products like @coworkerapp that dynamically route to the best open / closed models. average 80% cheaper while maintaining frontier AI
Now it looks like AI subscriptions were always running at a loss and the companies have to massively increase their costs to break even.
Was it all just a ploy to make people addicted to the usage of AI and the fear of being unproductive again the lever to extract the monies?
61
this is why the winner here will be products like @coworkerapp that allow you to intelligently route across any model while maintaining enterprise context
Deploying AI in enterprise is a mess right now.
We watched one company spend 8 months going in circles:
Month 1: Copilot (bundled in, seemed free)
Month 2: Rolled out ChatGPT to 20% of staff because Copilot underperformed
Month 3: Both tools sitting at ~20% adoption. Reassessing costs.
Month 4: Decided to go all-in on ChatGPT
Month 4-5: A rogue Claude user group quietly formed
Month 6: IT launched a formal Claude assessment
Month 7: Decided to switch the whole rollout to Claude
Month 8: ChatGPT Codex dropped. IT is now running another cost review...
This landscape will continue to change.
Enterprise AI adoption is not a procurement problem. It is a change management problem... And most companies are solving the wrong one.
1
118
Companies are realizing that 95% of their workloads can be accomplished with 80% cheaper models..
Anthropic faces AI spending backlash before IPO axios.com/2026/06/02/anthrop…
1
54
