Pinned Tweet
Great chat with @pk_iv on scaling agents for knowledge work
Recap of my time building @codegen, our approach to building the ultimate context engine at ClickUp, and unleashing agents on the open web via browser use.
Incredibly long on browserbase 🚀
Your agents are blocked by data, not models.
Last week I talked to @clickup's Head of AI @mathemagic1an, about why a dense context graph beats a smarter model every time. Their new Brain² Browser product uses @browserbase to fill forms, grab downloads, and bring external data into the platform.
07:45 What refactors taught clickup about narrow ai use cases
10:39 The biggest problem stopping companies from operationalizing ai
11:19 Why a dense context graph beats siloed data
12:09 Evals first: building the ai stack as a context engine
15:21 Skills as memory: clickup's answer to compaction
19:26 "I think we're close": are humans done writing code?
21:59 Ultracoding and the $120k bun to rust rewrite
24:13 What's actually stopping ai from curing cancer
26:30 Teaching an agent to order a pizza with just a browser
Watch the full episode of Navigators on YouTube, Apple Podcasts, or Spotify.
Agent browser use now live on ClickUp 👏
Some incredible applications emerging, feels like a true "unhobbling" moment.
Ordered a pizza live on an all-company demo and it nailed it. One of the more thrilling moments of my professional career.
Your AI can click through websites now. 🖱️
Brain² Browser fills forms, grabs downloads, and brings the results into ClickUp, even on sites with no integration. Built with @browserbase.
Our first test? Venues for our holiday party. 🥴
Rolling out to eligible Workspaces where enabled.
The future of HCI:
Omni models directly regressing onto UI pixels, locked in a full-bandwidth duplex interaction with the user
Within two years this will run efficiently on-device, calling out to cloud-based "oracle" models for high-value tokens.
Congrats to Gemini team 👏
Jay Hack retweeted
Today we're launching cve.now 🚨
Yesterday was a rough day for security. php2shell, LiteLLM, Railway... How are we supposed to stay on top of all of this?
cve.now aggregates 4 vulnerability sources into one free feed and uses AI to explain each one in plain English. No login.
cve.now
"Palantir for booking meetings"
What are we doing here guys
Please
F your AI b2b SaaS.
We just launched Palantir for booking meetings.
The last 14 days, companies like @MonacoGTM , @Nowadaysai, @Rippling , @salesforce , @marketercom, @posthog , and @retellai spent $100k+ with us on literal cakes.
From today, Daymaker finds your prospects, reads their entire digital footprint, buys the one gift that makes their day, and gets you the meeting.
Comment your name and I'll reply with everything Daymaker finds on you and the gift it would send you. If you dare. 🎁
A pleasure speaking on a panel with Anthropic @zkevinbai on the future of knowledge work and deployment of AI
Was on a panel with the one and only @mathemagic1an !! So cool to get into such a forward-thinking discussion on the future of technology, the future of software engineering, and the future of work!
Jay Hack retweeted
Today, we're launching Sazabi version 1.0.
The reality of shipping software as a team in the age of AI is actually crazy.
Agents merging their own PRs. Hundreds of deployments per day. A codebase that is constantly changing. Junior developers with 10x or 100x leverage. 😵💫
I spent the last year building a safety harness for teams that ship with AI. It's called @sazabi.
The premise is simple:
1. You send us logs.
2. We detect anomalies, investigate them, and send you back rich, actionable alerts about real problems in your app.
Sazabi helps you find issues before your users notice.
No monitors or noisy alerts. Just pure signal.
And today we're making it available for anyone to use.
The benefits of this are most apparent in chaotic, live environments
As we scale up the number of coding agents, expect environments to become far more dynamic and therefore unpredictable without heavy monitoring.
Spidermonkey delegates this to the subconscious.
Traditional coding evals like SWE bench focus on isolated tasks and do not capture the benefit of this approach
It improves token efficiency here for cheaper models (Luna) but does not help Opus
Also clear that this would require further tuning to hit max usefulness.
I think this cognitive architecture applies to agents beyond code.
Run decision models in a loop to process "early" features of your env and offload low-level computation from exec function.
Full implementation here:
github.com/jayhack/spidermon…
Humans rely on system 1 to navigate their environment - perception, reflexes, etc.
What if code agents did the same?
The idea: continually sample signals from your machine, then use decision models to identify and escalate events that matter
Is pure math just a jobs program?
Excellent read below to get a temperature check on math academic community. Variety of perspectives represented - a lot of feelings and anger.
Notably many of the critics cast it as a zero-sum competitive game with the labs, as though the point of math was to bring glory to mathematicians.
I could not disagree more - point of math is to reveal structure of reality.
We will have to redefine the role mathematicians play in society and this is sobering/scary. Definitely not the last time this happens for a profession.
But this mindset of math-as-jobs-program is near-sighted demonstrated lack of imagination. Many paths forward that increase prominence of pure math in our daily lives - demand for math talent will increase, similar to other Jevon’s paradox scenarios.
Had OpenAI not released these as an open GitHub repo, they would be accused of gate keeping knowledge. If someone invented a room-temp superconductor but kept it hidden from the world for fear of materials scientists losing status, we’d be horrified - this is no different.
Well worth reading: proofsandprompts.com/2026/10…
Jay Hack retweeted
ClickUp Brain is doing the thing everyone says agents will do.
Point it at your workspace and it writes code, runs it, hands back finished reports, decks, dashboards.
Not just answering questions. Doing the work.
ClickUp’s Brain² runs every frontier model with full context of your company, writing and running code to build reports, decks, and dashboards across all teams.
That code touches sensitive workspace data, has to start instantly mid-conversation, and has to scale to millions of users. That’s why @ClickUp built its execution layer on E2B: every sandbox is its own microVM, starts from a prebuilt template with tools ready, and scales on demand.
Enterprise-ready, from SOC 2 to HIPAA to full isolation.
The sandbox is a primitive for modern agent engineering
We're investing heavily - enables better artifact generation, data analysis, scaled agent context management via code mode, and much more
Will grow as a % of compute over time as agents expand in autonomy/responsibility
ClickUp’s Brain² runs every frontier model with full context of your company, writing and running code to build reports, decks, and dashboards across all teams.
That code touches sensitive workspace data, has to start instantly mid-conversation, and has to scale to millions of users. That’s why @ClickUp built its execution layer on E2B: every sandbox is its own microVM, starts from a prebuilt template with tools ready, and scales on demand.
Enterprise-ready, from SOC 2 to HIPAA to full isolation.
Partial progress on 3 millenium prize problems: Riemann, Hodge and BSD
We’re releasing a broad range of new mathematical results produced by an internal frontier model.
We’ve been consulting with the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study, and we have drawn on their advice and public recommendations to inform how we release these results.
github.com/openai/math
