Ehsan Nour (@thisisehsan) header
Ehsan Nour (@thisisehsan) profile picture

Designer using twitter. Seeking responsible tech. Wrote https://teapawt.net/t.co/dGxTJFdyxb Design @imbue_ai

Oakland, CA
Joined October 2013
Photo by Ehsan Nour (@thisisehsan)Photo by Ehsan Nour (@thisisehsan)Photo by Ehsan Nour (@thisisehsan)Photo by Ehsan Nour (@thisisehsan)Photo by Ehsan Nour (@thisisehsan)Photo by Ehsan Nour (@thisisehsan)Photo by Ehsan Nour (@thisisehsan)Photo by Ehsan Nour (@thisisehsan)Photo by Ehsan Nour (@thisisehsan)Photo by Ehsan Nour (@thisisehsan)
If you are: 1. using coding agents (or want to) 2. have appetite to try out a new agent orchestrator (ADE) 3. and willing to report at least 10 bugs hit me up, have something super cool for you to try out
1
2
4
298
as a designer who didn't used to code (beyond CSS), this transition for me is wild 🤯 in the last week, I designed and shipped a notes panel, @ mention system, and CMD K palette (just designing CMD K by hand would have taken me the same amount of time)
1
4
170
the pain point is that I now understand CI hell (even though agents make this much less of a problem than it used to be) and engineers who walk around with their computer open 🤦
4
83
Ehsan Nour retweeted
Deep learning works extraordinarily well. And we still largely don't know why. A new paper from @learning_mech, @KuninDaniel, and 12 co-authors argues that a scientific theory of deep learning is emerging, and coins a name for the emerging field: learning mechanics. We sat down with Jamie and Dan on Generally Intelligent to talk about what a physics of deep learning would actually look like, why now, and what's left to figure out. 3:05 Learning mechanics as the physics to mechanistic interpretability's biology 4:13 Why deep learning needs a theory 7:07 Why deep learning is uniquely hard to engineer 12:11 How a week in the woods became a paper 25:59 The barrier to theory isn't opacity, but complexity 36:26 Deep learning's first gas law 47:22 Why more particles makes the problem easier 56:22 The discretization hypothesis 1:01:50 The strongest signal that a compact theory exists 1:05:07 The Platonic Representation Hypothesis 1:15:41 Why learning mechanics and mech interp need each other 1:25:29 Theory as safety infrastructure
1
29
8
133
18,296
Ehsan Nour retweeted
Yesterday we launched Bouncer to put users in control of their Twitter feed. Today, we open sourced the Chrome extension. We believe software should align with your interests, not the company that built it. We built Bouncer for us, now go make it yours.
20
37
15
491
5,494,115
Ehsan Nour retweeted
Yesterday we launched mngr, an open-source CLI for running 100s of parallel agents. Today, we're releasing a detailed case study of how we're using mngr to run and improve itself, by testing its own demo script. Here's how it works:
3
10
1
84
5,268
Ehsan Nour retweeted
Introducing Vet: catch your coding agent’s mistakes Open source, fast, local. Zero telemetry Can review full PRs and it tells us when Claude/Codex/OpenCode lies about passing tests it never ran 👿 We use it daily and it saves tons of headache
21
25
8
282
586,272
Ehsan Nour retweeted
Your coding agent may be lying to you. You ask it to write tests. It says they pass. It never ran them. You ask for a feature. It hits a wall and silently swaps in fake data. We built Vet to fix this. It’s open source. Get the code below.
20
24
8
354
142,985
Ehsan Nour retweeted
220 employees of OpenAI and Google Deepmind have signed this letter urging their leaders to stand with Anthropic. That leaves... ~8780 to go. Friends at OpenAI and GDM, please join them! notdivided.org
45
338
29
1,575
53,457
Excited to share what we've been working on! We barely missed out on having an ARC-AGI-2 public high score (congrats @BingBongBrent and @bankminer78 ), but I'm most excited about our underlying approach: Our open-ended evolutionary method can optimize agentic systems end to end.
Today we’re open sourcing Evolver, a near-universal optimizer for code and text. While benchmarking we achieved SOTA (95%) on ARC-AGI-2 (last week that is 😆) and 3x’d performance of the best open model, reaching GPT-5.2-level performance.
1
3
1
36
5,274
Imbue research team putting in the work 💪
Today we’re open sourcing Evolver, a near-universal optimizer for code and text. While benchmarking we achieved SOTA (95%) on ARC-AGI-2 (last week that is 😆) and 3x’d performance of the best open model, reaching GPT-5.2-level performance.
5
223
I love that Imbue is hosting more and more events
This Post is from an account that no longer exists. Learn more
1
3
191
Had the honor to host a group of 30 design leaders, founders, and product folks at @imbue_ai this week and one thing is clear: I enjoyed myself a lot! Ping me if you want to join the next one
2
2
12
736
What conversations happen if you gather product leaders from companies like Anthropic, Figma, Gamma, and Stripe for dinner and a fireside chat with @joulee and @kanjun? Here were the top 3 highlights: 1) The roles of engineer, designer, and PM will converge, but taste will be the differentiator. Taste often seems ambiguous, but the core of it is understanding your user's needs. Traditionally, product managers have overseen coordination between designers and engineers. But as prototyping becomes easier than ever, ideas can now be evaluated through demos made in an afternoon. 2) You probably don't need more than 3 people to ship a successful product. Small startup teams have the advantage of decision-making speed, and AI expedites this even further. In January, we leaned into this at Imbue: we reorganized into multiple small project teams, each with a directly responsible individual (DRI) taking ownership of outcomes. Four weeks in, we're preparing to launch multiple end-to-end products. Our engineers now have the autonomy of founders, and are more energized than ever. 3) To preserve craft while balancing the speed of AI, carve out more time to think. Execution is no longer the moat; ideas are. While engineers have 10x-ed with AI, a designer's main challenge was never the speed of rounding corners in Figma, but the thinking behind making products useful and beautiful. The more designers are able to reflect about the user's needs, the stronger their intuition will be; increasing the conviction of their design decisions and subsequently the impact of their products.
1
2
179
A designer's main challenge was never the speed of rounding corners in Figma, but the thinking behind making products useful and beautiful. Execution is no longer the moat; ideas are.
2
90
Had a lot of fun teaching my co-workers @imbue_ai about Figma yesterday Their minds were particularly blown when I showed off some of the smart alignment tools. Things may or may not have kept escalating as the lesson went on, nearly had a revolt on my hands by the end 😂
1
7
685
If you write code, you are going to want to check out what I have been working on for the last year. Sculptor is such a seamless way to write code with AI and integrate it with whatever workflow you are already using.
Sculptor: the missing UI for Claude Code 🎨 Imagine running 5 Claudes in parallel, safely in containers, while you stay in flow. Then bring their work straight into your IDE to test/edit together. This is how one developer ships like a team. Try it with Sonnet 4.5!
14
619
Ehsan Nour retweeted
Generally Intelligent is back—and now in video! Our first guest is our very own @mattboulos, who leads policy at Imbue. He and @kanjun discuss: - how AI shifts power - lawless digital spaces - bottom-up vs. top-down automation and more. Full episode and transcript in thread:
1
6
3
23
10,892
Trying to calibrate for a brand design hire... I am looking for examples of exceptional brand design portfolios. Any favorites?
1
1
2
500
Have you ever had a client say the words "I'm not a designer, but..." ? Get ready to practice the hardest skill in design. Link to my full #Config2025 talk below.
14
18
3
350
37,151