Speed is a commodity, judgement isn't
Cerebras announces 750 tokens per second. Alibaba releases open weights for a model that, they say, beats its predecessor at code and office work. Apple wants Siri to read the news with judgement.
Three pieces of news, one signal: the marginal cost of intelligence is collapsing. And when something becomes cheap and abundant, it stops being an advantage.
So the operational play, as I see it, is this: stop competing at the model layer. You're not going to win there. Cerebras, NVIDIA, Alibaba and OpenAI will. You compete one layer up. The layer of "which real problem, for which specific customer, do I solve with this".
A quick framework to think it through:
- Which process in my business (or my customer's) still costs human hours and doesn't care who does it?
- What would happen if that process cost cents and ran in seconds?
- What new product or service can I build on top of that new reality?
If you answer those three well, it doesn't matter whether GPT-6, 7 or 8 ships. Your thesis holds.
If the only thing you know how to do is plug an API into a nice prompt, any kid with Cursor will eat you in a weekend.
What I learned
My take this week: tech hype is an emotional trap for founders.
Every time a new model, a new chip or a new announcement drops, there's a little voice telling you "maybe you should rethink the strategy". And it's almost always lying.
You rethink strategy when the customer's problem changes, not when the supplier's toy changes.
I have to remind myself of this every week, because the noise is constant and so is the temptation to chase the shiny thing. In the end, the boring work (talking to customers, understanding their pain, iterating the product) is what moves the needle. The rest is entertainment for founders.
On the radar
- Cerebras hits 750 tokens per second with a 900,000-core chip. Beautiful. And equally irrelevant if you don't know what to build on top.
- Alibaba releases open weights for Qwen3.8-27B with a native 262K context. Open source pushing hard. Depending on a single closed provider is starting to smell like a lazy decision.
- Apple wants Siri to deliver news with AI and deals with publishers. They're late, but when Apple arrives, the distribution game changes. Watch out if your product lives on the phone.
- Expansión launches a weekly barometer of the most in-demand professional skills. Translation: the job market is being rewritten in real time. If you don't learn to use AI day to day, it isn't that you fall behind, it's that you disappear.
The actionable advice
Take your week. Note the 5 tasks you spend the most hours on. Next to each one, write: "could a current model do this for under a euro?". If the answer is yes for three or more, you have a problem with how you allocate your time. And probably a product opportunity right under your nose.
Behind VIVIO
Applied to VIVIO, my view is clear: in influencer marketing, the fastest model doesn't win. The winner is whoever genuinely understands the brand's problem and the creator's problem.
A brand doesn't pay you for having the best prompt. It pays you for saving it hours of research, negotiation, results analysis and blind decisions. A creator doesn't stay on your platform because you use the latest model, but because you solve getting paid, understanding their value and working with serious brands.
My product bet: AI is infrastructure, not a value proposition. The value is in the judgement with which you apply it to the industry's real workflow.
This is a version of what I send every week, unfiltered, in my newsletter "Eso no fue suerte". If you want the rest, subscribe for free.