Pime Pardo

over 1 year ago

And I think it’s such a lucky accident, having been born, that we’re almost obliged to pay attention. Mark Strand.

about 9 hours ago

Buena @donjulio.

pimepardo retweeted

Tess

@tess_us

1 day ago

Launched today by Anthropic, Claude Opus 5.5 is already live on Tess. We put it to the test with our YouTube video analysis tool.

Co-founder at @platan_ventures // family + work 💯

1 day ago

Waaay tooo coool

Federico Antoni

@federicoantoni

1 day ago

Found this little time capsule from my previous life in fashion and beauty. Santa Monica, many years ago, shooting with @AnneV. I’m somewhere in the background, with long hair :)

433

222

Who to follow

2 days ago

Must read 👇🏽

Rica Barros

@souorica

2 days ago

Jev, explained for executives: what it actually means for your business I wanted to write this for founders, executives and operators trying to understand where Jev fits into their business. Not a technical breakdown. Just what it does, where it helps and what we’re seeing in our own tests. Last week, TypeSafe AI launched Jev, and it’s been getting a lot of attention. The company is led by Diogo Almeida, a former OpenAI researcher credited as one of the inventors behind ChatGPT, has $40M in funding, and got a TechCrunch headline about “thrilling developers.” The claim is that this is a new class of AI model, not just another model. In the demo, Jev made a decision in 0.1 seconds that took GPT-5.6 8.5 seconds, at roughly 1/50th of the cost. Those are big numbers. But what does that actually mean in practice? 1. What it is Jev is not a chatbot. It doesn’t write emails, summarize documents or brainstorm with you. It makes fast, structured decisions. You give it context and criteria you’ve defined beforehand, and it returns a yes/no or a choice from a list, with a probability attached. “Does this customer want a human?” → 92% yes. “Is this task actually complete?” → 31% yes. “Which of these 5 workflows should run?” → Workflow 3, 87%. A useful way to think about it is Kahneman’s System 1: fast, automatic responses. In that analogy, LLMs play the slower, more deliberate System 2 role. It’s a simplified comparison, but it helps explain where each one fits. 2. Why it matters For a business, the appeal comes down to three things. Speed. Responses take tens to hundreds of milliseconds rather than seconds. When you have agents making decisions at every step of a task, that adds up quickly. Cost. The pricing is a small fraction of even the cheapest LLMs. For decisions you’re making thousands or millions of times, that can change the economics. Calibration. This is the less obvious one, but it matters. Jev is trained so that its probabilities reflect how often it’s right. The goal is that, across decisions where it gives a 60% probability, it’s right about 60% of the time. That gives you a more useful basis for deciding when to let the model act and when to involve a human. Also, it can only choose from the options you give it. It can still choose the wrong answer, but it can’t invent a new one outside that list. 3. What it is not This doesn’t mean Jev can replace an LLM. It isn’t built to work through open-ended problems or reason through unfamiliar situations in the same way. Every possible answer needs to be defined beforehand. That limits what you can use it for. But for a lot of business tasks, a fixed set of answers is exactly what you need. 4. Where it gets interesting A lot of what companies use LLMs for today is simply choosing what happens next. Route this ticket. Pick this agent. Escalate or not. Approve or flag. Check whether the task is done. Decide where to click. We’re paying LLM prices, and waiting for LLM responses, for many decisions that may not need a generative model at all. With models like Jev, you can move some of those decisions to something faster, cheaper and better calibrated, while keeping the LLM for the parts that actually need reasoning or language. You don’t have to replace the whole workflow to get value from it. 5. What we’re doing at TESS We’ve been testing it. In our early internal tests, it was roughly 10× faster at one-third the cost of GPT, and it got every case we ran right. It was a small sample, not a production benchmark. I wouldn’t read that as “100% accuracy.” But it was enough for us to take it seriously and keep going. If those results hold up, we could move many of our background LLM calls to this type of model: agent routing, task routing, escalation and checking whether an agent actually finished the job. For customers, that could mean better decisions, fewer half-finished tasks and lower costs. The use case I’m most interested in is browser use. It’s one of the biggest token burners in agentic AI, and a lot of those calls involve deciding what to do next on a page. If we can move some of those navigation decisions to a model like this, browsing could become much faster and cheaper. The takeaway Not every step in an AI workflow needs an LLM. Sometimes you just need a reliable decision, quickly. That idea isn’t new. What’s interesting about Jev is whether it makes this approach practical and cheap enough to use at scale. From what we’ve tested so far, it’s worth paying attention to.

$souorica's tweet photo. Jev, explained for executives: what it actually means for your business I wanted to write this for founders, executives and operators trying to understand where Jev fits into their business. Not a technical breakdown. Just what it does, where it helps and what we’re seeing in our own tests. Last week, TypeSafe AI launched Jev, and it’s been getting a lot of attention. The company is led by Diogo Almeida, a former OpenAI researcher credited as one of the inventors behind ChatGPT, has $40M in funding, and got a TechCrunch headline about “thrilling developers.” The claim is that this is a new class of AI model, not just another model. In the demo, Jev made a decision in 0.1 seconds that took GPT-5.6 8.5 seconds, at roughly 1/50th of the cost. Those are big numbers. But what does that actually mean in practice? 1. What it is Jev is not a chatbot. It doesn’t write emails, summarize documents or brainstorm with you. It makes fast, structured decisions. You give it context and criteria you’ve defined beforehand, and it returns a yes/no or a choice from a list, with a probability attached. “Does this customer want a human?” → 92% yes. “Is this task actually complete?” → 31% yes. “Which of these 5 workflows should run?” → Workflow 3, 87%. A useful way to think about it is Kahneman’s System 1: fast, automatic responses. In that analogy, LLMs play the slower, more deliberate System 2 role. It’s a simplified comparison, but it helps explain where each one fits. 2. Why it matters For a business, the appeal comes down to three things. Speed. Responses take tens to hundreds of milliseconds rather than seconds. When you have agents making decisions at every step of a task, that adds up quickly. Cost. The pricing is a small fraction of even the cheapest LLMs. For decisions you’re making thousands or millions of times, that can change the economics. Calibration. This is the less obvious one, but it matters. Jev is trained so that its probabilities reflect how often it’s right. The goal is that, across decisions where it gives a 60% probability, it’s right about 60% of the time. That gives you a more useful basis for deciding when to let the model act and when to involve a human. Also, it can only choose from the options you give it. It can still choose the wrong answer, but it can’t invent a new one outside that list. 3. What it is not This doesn’t mean Jev can replace an LLM. It isn’t built to work through open-ended problems or reason through unfamiliar situations in the same way. Every possible answer needs to be defined beforehand. That limits what you can use it for. But for a lot of business tasks, a fixed set of answers is exactly what you need. 4. Where it gets interesting A lot of what companies use LLMs for today is simply choosing what happens next. Route this ticket. Pick this agent. Escalate or not. Approve or flag. Check whether the task is done. Decide where to click. We’re paying LLM prices, and waiting for LLM responses, for many decisions that may not need a generative model at all. With models like Jev, you can move some of those decisions to something faster, cheaper and better calibrated, while keeping the LLM for the parts that actually need reasoning or language. You don’t have to replace the whole workflow to get value from it. 5. What we’re doing at TESS We’ve been testing it. In our early internal tests, it was roughly 10× faster at one-third the cost of GPT, and it got every case we ran right. It was a small sample, not a production benchmark. I wouldn’t read that as “100% accuracy.” But it was enough for us to take it seriously and keep going. If those results hold up, we could move many of our background LLM calls to this type of model: agent routing, task routing, escalation and checking whether an agent actually finished the job. For customers, that could mean better decisions, fewer half-finished tasks and lower costs. The use case I’m most interested in is browser use. It’s one of the biggest token burners in agentic AI, and a lot of those calls involve deciding what to do next on a page. If we can move some of those navigation decisions to a model like this, browsing could become much faster and cheaper. The takeaway Not every step in an AI workflow needs an LLM. Sometimes you just need a reliable decision, quickly. That idea isn’t new. What’s interesting about Jev is whether it makes this approach practical and cheap enough to use at scale. From what we’ve tested so far, it’s worth paying attention to.$

226

110

8 days ago

- Figuring out what to build is more important now than ever. -

Josh Elman

@joshelman

10 days ago

https://t.co/8zuACcwHGR

982

110

617K

128

pimepardo retweeted

14 days ago

Why are traditional industries adopting AI agents faster than tech sectors? In this clip, @tess_us Co-founder CGO Milena Maia breaks down the main obstacle to AI adoption in enterprises, based on insights from founder & CEO Ricardo Barros for the State of AI report by HI Ventures. Discover why successful AI deployments start bottom-up, not through rigid IT rollouts.

287

16 days ago

- love and kindness is the new punk - John Cameron Mitchell

148

pimepardo retweeted

Y Combinator

@ycombinator

17 days ago

Harnesses often get dismissed as just scaffolding, just prompt engineering, and not real research. But that couldn't be farther from the truth. The same model weights that score 30% on ARC-AGI score 95% with a better harness. So we gathered a group of researchers and founders working at the frontier to do a deep dive into the state of harnesses. We cover how we got to this point, the case for making your harness as expressive as possible, and what YC learned building an agent for every employee in the company. 00:00 - @FrancoisChauba1: Why harnesses matter 04:27 - Building an auto-researcher by accident 07:13 - A five minute history of harnesses 13:56 - Self-improving harnesses 18:35 - @sethkarten: Prime Agent, a self-improving RLM harness 21:50 - Context as an L1, L2, L3 cache 24:51 - From Turing machine to von Neumann computer 28:33 - Messaging between agents 30:04 - ARC-AGI results 33:09 - Emulator Bench and GPU kernels 37:30 - @JonSaadFalcon: OpenJarvis, personal AI on personal devices 38:26 - How far behind are local models 39:21 - The five primitives of a personal AI stack 42:47 - Letting cloud models optimize your local stack 43:53 - 800x cheaper than the cloud 45:58 - @josh__france and @jbellregan: QM, YC's agent harness for work 47:29 - A history of YC's internal agents 49:24 - OpenClaw and a fleet of 50 agents 51:04 - Pulling the brain out of the sandbox 54:43 - Letting the agent choose its own sandbox and model 57:16 - The grind tool: budgets on goals 58:50 - Agents don't understand social context

183

444

490K

pimepardo retweeted

Tess

@tess_us

20 days ago

Most AI agents die in the IT backlog. The person who understands the process best usually sits in finance, in operations, in support. They can describe exactly what the agent should do. Then they open a ticket, and wait. Our founder & CEO, @souorica, shows you how Tess solve this, turning AI implementation into real business results.

pimepardo retweeted

21 days ago

Jean Marc, fundador de Konko AI, se une a Federico Antoni y Jimena Pardo para hablar sobre construir en la intersección entre la inteligencia artificial y la salud en Latinoamérica. Hablamos de: -Por qué el mayor reto de la salud es una oferta limitada frente a una demanda ilimitada. -Cómo la AI puede multiplicar la capacidad de médicos y clínicas. -Por qué el contexto del paciente y la interoperabilidad son fundamentales para el futuro de la salud. -Cómo construir una empresa implica tomar riesgos, fallar y aprender en el camino. Una conversación sobre AI, salud, emprendimiento y por qué nunca había habido un mejor momento para construir.

hi_ventures_'s tweet photo. Jean Marc, fundador de Konko AI, se une a Federico Antoni y Jimena Pardo para hablar sobre construir en la intersección entre la inteligencia artificial y la salud en Latinoamérica.

Hablamos de:
-Por qué el mayor reto de la salud es una oferta limitada frente a una demanda ilimitada.
-Cómo la AI puede multiplicar la capacidad de médicos y clínicas.
-Por qué el contexto del paciente y la interoperabilidad son fundamentales para el futuro de la salud.
-Cómo construir una empresa implica tomar riesgos, fallar y aprender en el camino.

Una conversación sobre AI, salud, emprendimiento y por qué nunca había habido un mejor momento para construir.

353

pimepardo retweeted

Autism Capital 🧩

@AutismCapital

27 days ago

Disappointed Katy Perry didn't make the list.

192

42K

pimepardo retweeted

29 days ago

AI can close Latin America’s productivity gap Automation presents a massive expansion opportunity for Brazil, Chile, Mexico, and Argentina — the room to grow is exactly where AI can make the biggest difference. Explore this and more in the 2026 State of AI Latin America. Full report in the comments.

hi_ventures_'s tweet photo. AI can close Latin America’s productivity gap

Automation presents a massive expansion opportunity for Brazil, Chile, Mexico, and Argentina — the room to grow is exactly where AI can make the biggest difference.

Explore this and more in the 2026 State of AI Latin America. Full report in the comments.

298

pimepardo retweeted

Brad Flora

@bradflora

about 1 month ago

You are not the main character. Your user is the main character.

850

106

31K

about 1 month ago

In Mexico, if your startup isn’t growing, you can just cast a spell on your customers… take that AI!👇🏾✨

162

about 1 month ago

where do I join the waitlist?

Federico Antoni

@federicoantoni

about 1 month ago

Trying to put my @TownAI agent into this device… …wish me luck :)

491

270

about 1 month ago

Esto es @fintual 👇🏽

283

about 1 month ago

A garota de Ipanema, em versão relatório de IA....