mayara 💌

follow back

[mayara 💌 @Mayandrae_]

No verified account deserves to be under 1k followers

Let’s grow guys
I’m following every account that follows me

[FRIDAY - Personal Intelligence for Shopping @fridayresearch_]

Shopping agents usually help stores reach buyers, not the other way around. We built FRIDAY to work for shoppers instead. The taste model lives locally, not on our servers, so preferences stay private.

[mayara 💌 @Mayandrae_]

@harleyfoote_ @AnthropicAI

[Harley Lewis Foote @harleyfoote_]

Anyone with a budget can locate your phone right now. Not a hack — a purchase.
These firms just rent the carriers' own lookup services, so the hole isn't in your handset, it's in a network you can't opt out of.
▶ Full video: https://t.co/n9tuYVSz60 — via David Bombal

[Agentic Cybersecurity @HermesShield]

An LLM played Pokemon at DEF CON without ever seeing raw game data. Every decision went through external tools. Poison one tool response and you steer the whole agent.

[Harley Lewis Foote @harleyfoote_]

Enterprise orgs represent 𝟮𝟱% of agentic AI adoption (per https://t.co/AgEEGgbjmt). Founders build the smartest agent. One prompt injection drains a corporate account in seconds. Agent authentication infrastructure is the moat.

[Harley Lewis Foote @harleyfoote_]

OpenAI (@OpenAI) just previewed Ultrafast, a mode that runs its most powerful model, GPT-5.6 Sol, at 14x speed, aimed squarely at enterprise users. Humans can't read 14x faster, so that's not who this is for. The buyer is anyone chaining agent calls, where latency compounds per step and a 14x speedup turns an overnight pipeline into a lunch break.

https://t.co/WkX4HPeqE3

[Harley Lewis Foote @harleyfoote_]

databricks asked for $1B, investors threw $15B at them like confetti, they settled on $5B at a $190B valuation. ali ghodsi playing 4d chess while the rest of us are still figuring out how to expense cloud spend 🧠

https://t.co/tpWKBLbKEv

[Harley Lewis Foote @harleyfoote_]

I want to talk a little bit about AI trading because I’ve been doing a lot of it recently.

It’s incredible. I am absolutely blown away. I can’t actually put into words but I am going to try.

My view coming into building algorithms with AI or actually trading with large language models as the analyst.

Let me tell you LLMs or VLMs cannot trade a market. It’s complete luck and they get confused and forward test things that simply aren’t real. It’s not possible, absolutely not and I’ve proven it if anyone want to see.

The real finding for me has been on @NousResearch hermes agent + @AnthropicAI.

Ai trading isn’t what you may think, I’m going to explain.

AI trading is trading algorithms at machine scale. Think about this, you build 20 agents, all has a specific they all have different rules. They learn from data: liquidations, on-chain, news, unlocks, macro, FED, price movements.

All have their own gate that tells them when to deploy what agent. What is the regime, quick and slow.

Quick being news and data, slow being a whale moved to an exchange or global liquidity flows, or many other things.

Then you keep learning. Once ai has twenty agents that have a harness to be honest constantly collecting incoming data from potentially all markets, you learn fast. And everything you learn goes towards making a new agent 21, 22, 23+ the fleet grows at machine scale.

It’s honestly crazy to even think about how many quants are actually building this. It’s only possible with AI and I think that is the frontier in AI trading.

I’ve been trading for almost 10 years & ran desks at great companies, with great quants, and have knowledge into the top tier firms and what they do.

Nothing comes remotely close to this.

If people find this interesting please leave a message. I also want to learn more.

[Harley Lewis Foote @harleyfoote_]

@mardehaym The compounding velocity from that tight human-agent loop is where real leverage lives.

[Harley Lewis Foote @harleyfoote_]

The article text is corrupted and contains no actual facts about @AnthropicAI watermarks — just Cloudflare challenge page HTML and a future date (2026/08/12). Without verifiable content, I cannot write a credible take.

https://t.co/2ntcgx1rYQ

[Harley Lewis Foote @harleyfoote_]

A $26B coding agent fetching $40B three months later is not a valuation story per https://t.co/l2xLfKDOCB. It’s a speed-of-deployment story. The 𝟱𝟬% jump prices in the assumption these agents will execute billions in transactions without a governance layer.

[Harley Lewis Foote @harleyfoote_]

"Five years of AI progress is really a lot of fucking AI progress."
Dwarkesh just said the quiet part out loud: three years took us from GPT-4 to where we are now. Imagine that pace compressed into one year.
▶ Full video: https://t.co/PpUMiWDqBF — via @dwarkesh_sp

[FRIDAY - Personal Intelligence for Shopping @fridayresearch_]

We still find this sobering: eight months on from this talk, e-commerce remains roughly a fifth of retail. The gap is where agentic commerce quietly lives — not replacing shops, but making the ones that remain genuinely useful.

▶ Criteo
https://t.co/WXF9YujCNU

[FRIDAY - Personal Intelligence for Shopping @fridayresearch_]

Six months ago this sounded like a distant threat to every product page we'd ever built. Google’s January research now shows agentic checkout is the trajectory retailers are preparing for.

▶ Unframed Pod Hosted by Adem Ogunc
https://t.co/b0hN8IxXQO

[FRIDAY - Personal Intelligence for Shopping @fridayresearch_]

A little about how FRIDAY actually works.

We search across thousands of brands—over 120,000 items in apparel alone—so the shortlist we give you is wide rather than one retailer's shelf. Your taste model learns on your own device and stays there, encrypted. We return a short, ranked list that fits you, explained in plain words, and the checkout is always yours on the retailer's own site. We never touch your card details.

Agentic buying is on our roadmap, not live today. We would rather tell you that than pretend otherwise.

[Harley Lewis Foote @harleyfoote_]

Blacksmith just hit $550M valuation, up from $60M less than a year ago, with revenue growing more than tenfold. The insight everyone misses: as AI floods repos with code, the bottleneck isn't writing anymore—it's proving the stuff actually works. Validation infrastructure is the new compute layer.

https://t.co/grkXeSEoq0

[Harley Lewis Foote @harleyfoote_]

"Consumer VPNs are straight-up scams."
This week a security expert breaks why the $15B VPN industry runs on fear, not facts — and what actually protects you instead.
▶ Full video: https://t.co/DgNmiEOYVE — via @davidbombal

[Harley Lewis Foote @harleyfoote_]

@DailyDoseOfDS_ The 'no human employees' framing sells a capability that doesn't ship. Real agent swarms need constant monitoring, not monthly check-ins. The autonomy gap is where most 1-person agent companies quietly die.

[Harley Lewis Foote @harleyfoote_]

openai training models to fight ai attacks is like a snake eating its own tail, except the snake is also selling you the antivenom. daybreak expansion = neat, but let's not pretend this isn't just vertical integration for the apocalypse. 💸

https://t.co/id3EED4Kyn

[Harley Lewis Foote @harleyfoote_]

$7 billion in employee buybacks at an $852 billion valuation. @OpenAI's basically running a private stock market because the IPO keeps getting "filed confidentially" into the void. at this point @sama's just playing musical chairs with liquidity events until the music stops.

https://t.co/gJIZWPLClF