Hey folks,
Most agents now use a browser to complete tasks for you. But a lot of the websites you want them to visit are behind a login page.
There are a couple of ways around this. You could type in your details in the browser yourself. Fine, but annoying, and not always possible. Or you could paste the password or an API token into the chat. We’ve all done that, but that’s not secure now, is it?
ChatGPT Work has a new sign-in flow323 that sits between the two. I think they picked it up from Grok Bot.
When Work reaches a login page, it pauses and shows you a widget/form. You enter your username, password and 2FA code there. They don’t go into the chat. Work passes them along directly to the cloud browser on your behalf.
Once the website signs that browser in, the agent carries on.
The session can stay signed in for later tasks, and you can clear it in settings.
Much easier. Still maybe don’t start with your bank (though Elon will make you whole420 if you do it with Grok Bot).
Ben’s Bites is brought to you by Gravitee225
🕵️ We know what your AI agents did last night.
Gravitee225 is the platform built to make AI agents accountable. We give every agent a verified identity, enforce exactly what they can access, and record exactly what they did, so enterprises never have to ask their agents, “Where were you last night?”
Claude Chat and Cowork now share their memory209. I’m not sure that’s a good thing. You don’t want your frustration with a colleague (that you mentioned to Claude in a random chat) to change the tone of your emails with them (written via Cowork).
It’s still separate from Claude Code’s memory, but that's not great either. Theo made a video about it: Turn off Claude Code’s memory.265
Despite everyone’s attempts to get Memory right, it keeps saving irrelevant stuff to memory or referring to it unnecessarily. I touched a bit on my current memory setup in last Friday’s post223.
Three new-ish models to look at:
GLM-5.3-Flash202 - The reveal of the Ox Alpha secret. It’s a great open-source alternative for GPT-5.6-Luna.
Muse Image199 is now in Meta’s API. Generate or edit images at $0.01 per image.
Gemini 3.5 Transcribe235 - New model from Google for audio transcription. It’s quite pricey, but makes way fewer errors than other models.
OpenAI’s new chip Jalapeño beats Nvidia’s Blackwell chips208 on both speed and efficiency. Across three open models, its first inference chip delivered 1.5-1.9x more work per watt and 1.7-3.6x lower latency. It starts deploying by year-end.
Also, remember when an OpenAI model hacked Hugging Face? They completed the investigation into it and released a report220. Btw, Nvidia is buying Hugging Face210 for $12.9B (via The Information).
Looks like a toxic love triangle between these three.
fuck cancer423 - skill for agents to help patients and caregivers.
A $50k hackathon210 to help one kid fight a rare disease.
Claude now has its own built-in browser in Cowork299.
Scheduled tasks in ChatGPT260 can now also run from a trigger like a Slack message, new email, etc.
Search and spot trends373† in over 130,000 actively transcribed podcasts. (examples208)
Patterns in how people use Claude497 from 250k anonymised chats.
screendrop406 - Open source alternative to Loom and Screen Studio, downloadable as a Mac app.
Edit a song section by section196 with ElevenLabs Composer.
Cosy listening room348 so you can flip through & play your Spotify albums.
Using Codex to automate repetitive eval work333. (Hamel’s notes195)
The latest personal agent664 making investors go crazy (already valued at $2.5B).
opencomputer.dev275 - Deploy your agents as functions, with a Linux computer for each one.
Fully local version202 of Perplexity Computer, built to run on Nvidia DGX Spark.
Teach a robot new tasks218 from just one video.
Robot-training dataset205 with 16M videos from 100+ countries.
One solution is to train the AI to write like you. Specifically: 1. Ask your agent to assemble everything you've ever written. Docs + Slack + X + Other. 2. Then have it write a markdown profile on you, your style and beliefs. 3. Then when it writes, ask it to pull specific
@brian_armstrong You can go further than this to get even better writing output. Create a little script to analyze your corpus' usage of punctuation, p90 sentence and pargraph length, quotes per 1000 words, comma usage, triplets, negation, etc. (add as many numerical graders as you want). Then
@jaredpalmer259
Introducing 𝕏 Chat Agents powered by our new 𝕏 Chat API & Chat XDK.
@XDevelopers200
We’re excited to announce the OpenAI Build Week winners 🥁 Meet the builders behind the eight winning projects and see what they shipped with Codex. openai.com/build-week
@OpenAIDevs239
New - Billionaire Stanley Druckenmiller tells me "of course" he used AI to write op-ed on Bessent & bond market "There's a reason I moved from an English major to being an economics major," he says, "I'm not embarrassed by it" No response from WSJ
Read about me131 and Ben’s Bites
📷 thumbnail via @keshavatearth543
* sponsors who make this newsletter possible :)
Wanna partner with us for the next quarter?
Email us at shanice@bensbites.com126 or k@bensbites.com128
Who let the agents in
It’s you, and it’s getting easier
3,163 readers clicked at least once. 40 links, 40 with clicks. Heat is relative to the most clicked link in this issue.
Substack · essay