Marco Rodrigues

When potential clients reach out about integrating AI agents, I give them two paths: 1. Build from scratch. 2. Hermes Agent by @NousResearch. So far, every single client has picked option two.

@NoemiTitarenco Same, memory is overrated!

I only use 3-4 skills. I don't use any memory system. I hate context pollution. Am I the only one?

@eth_systems Looks promising, but please change the font of the website 🙂

Today we're launching EthSystems. We build confidential systems for institutional Ethereum. Institutions want to use Ethereum, but one of the biggest problems is the lack of built-in, modular privacy tools. We were the Ethereum Foundation's Institutional Privacy Task Force (IPTF) for the past year. We had hundreds of conversations with central banks, regulators, tier-one banks, and asset managers, shipping open source work the whole time. Wall Street has found crypto as an asset class, but not yet as commercial infrastructure. Institutions want to run real flows on Ethereum: stablecoins, tokenized assets, settlement. These are businesses with billions of dollars on the line, and no bank will operate in full public view. On a public ledger, confidentiality is the hard part: each party to a transaction should see what it has a right to see, and nothing more. We have a year of proof of work: private bonds, confidential stablecoin transfers, private settlement across chains, the Ethereum Privacy Map, and more. All with protocol specs and security properties, at our website. We've spent a decade working on privacy in crypto. We know there's no silver bullet. Different use cases need different systems, each designed, specified, and hardened properly, and someone has to do that work. That's why EthSystems exists. We're an independent, for-profit company, backed by long-term Ethereum-aligned investors. This is a decade-long transition, and we aren't going anywhere. If you're an institution that wants to build on Ethereum, talk to us. We're hiring: BD in New York, protocol engineers, ops: join@ethsystems.org

@arthur0x That's because you never lived in Spain 😅

Portugal easily has the worst bureaucracy I have ever experienced, extremely dysfunctional, slow, and unprofessional. The country and people are lovely but man the civil service is worse than most third-world countries.

@OpenRouter This new look burns my eyes.

OpenRouter has a new look. As we’ve grown into the infrastructure powering the intelligence era, our brand needed to evolve too. Built on clarity, precision, and purpose. For developers. For enterprises. For everyone turning intelligence into impact.

@levelsio Buy some whole milk. Boil it. Let it cool down a bit. Add bacteria to it. Wait for 12 to 24h Eat a proper yogurt.

Most greek yogurt is dogshit This is Oikos greek yogurt in Portugal It's 3 grams of protein!!! 3%!!! Insanely low You're mostly eating fat and calories!

😅

I was promised a burning Paris. What is this disgusting relatively civilized behavior?

@DavidOndrej1 Use Grok

Codex is no longer a "better deal" than Claude Code...

Is Grok 4.5 on @cursor_ai still not available in Europe? I need to use it with my OpenRouter key. Very disappointed.

@DurovPD What is this? Polymarket on Telegram?

$2000 to 2 people who predicts correct score. ends in 24 hrs

@OliseM17 Nice, filmed in Ouarzazate?

France vs Morocco 🇫🇷 x 🇲🇦

@AlexFinn What are Opus 5 and GPT 6? 🤣

We are living in the greatest 30 day time period in technology history In the month of July all of these have been released/are expected to be released: • Fable 5: smartest AI model ever • Muse Spark 1.1: Meta back in the race • Opus 5: affordable version of Fable • Grok 4.5: affordable version of Opus + Elon’s back • GPT 5.6: Affordable, lightning fast, better than Fable in some ways • GPT 6: Goblin mode • GPT Work: Amazing new way to work The way you work 30 days from now will be dramatically different than the way you work today You will be significantly more productive with significantly less effort required Thank your God he/she decided you should be alive right now

Stop using queues. Start using: - /goal - /steer - /background - /learn And many more. With the Hermes Agent by @NousResearch.

I don't know about the models. But OpenAI is killing it with ads.

So now Luna or Grok 4.5? 🤔

@rezoundous Grok 4.5 is the "cheapest frontier"!

Grok might just become the people's AI

@Hesamation Grok 4.5 is the "cheapest frontier"!

wtf Grok

Frontier AI just got half as expensive. Grok 4.5 is the "cheapest frontier". Fable 5, Claude Opus, and GPT-5.5 still win many of the intelligence rankings. If you only look at the top of the benchmark tables, Grok 4.5 is usually behind them. But only slightly. Grok 4.5 scores 64.7% on SWE-Bench Pro. Opus is still ahead there, around 69.2%. It reaches 83.3% on Terminal-Bench 2.1, close to GPT-5.5. It runs at around 80 tokens per second. And according to Artificial Analysis, it lands around fourth overall, behind Fable 5, Opus, and GPT-5.5. So why does this matter? Because Grok 4.5 costs $2 per million input tokens and $6 per million output tokens. That is much cheaper than the Western models it is competing with. For a freelancer, startup, or small team, this changes the calculation completely. Before, the choice was often: > Use the best model and burn money fast. > Use a cheaper model and accept weaker results. > Use Chinese models like DeepSeek or Qwen because the price gap was impossible to ignore. Grok 4.5 is the first Western frontier model that comes close enough to compete on price while still performing near the top models. That makes AI more accessible. > More people can run agents for longer. > More people can test ideas without watching the bill every minute. > More people can use near-Opus quality without paying Opus prices. Imagine what happens when the whole market has to price against this. Because once people get 90% of the performance for a fraction of the cost, they don't easily go back.

One thing I'm liking about Grok 4.5 is that it often asks for confirmation before applying changes.

Grok 4.5 is very good, with Composer 2.5 speed