edwin emmanuel roy (@mr_edvi) header
edwin emmanuel roy (@mr_edvi) profile picture

I fix the product before the screens exist. UX for startups that can’t afford a rebuild. https://teapawt.org/t.co/ebijEk1L4O · Kerala

Trivandrum, India
Joined February 2021
Photo by edwin emmanuel roy (@mr_edvi)Photo by edwin emmanuel roy (@mr_edvi)Photo by edwin emmanuel roy (@mr_edvi)Photo by edwin emmanuel roy (@mr_edvi)Photo by edwin emmanuel roy (@mr_edvi)Photo by edwin emmanuel roy (@mr_edvi)Photo by edwin emmanuel roy (@mr_edvi)Photo by edwin emmanuel roy (@mr_edvi)Photo by edwin emmanuel roy (@mr_edvi)Photo by edwin emmanuel roy (@mr_edvi)
To the founders transforming legacy products into agentic experiences: Why will your users trust a stranger to fly them 35,000 feet above the ground, but hesitate to let an AI send a two-line email? That's the question I'd start with. Because the challenge with agentic UX isn't simply: “How do we give AI more autonomy?” It's: “How do we make humans comfortable delegating control to it?” For decades, UX looked like: Human → Interface → Action Agentic AI changes that to: Human → Intent → AI → Decision → Action The interface is no longer just helping users do things. The system is increasingly doing things for them. And that changes everything about how we think about trust, control and agency. So if you're taking a legacy product, adding an AI agent on top, and calling it an agentic experience, you might be solving the wrong problem. I went down a rabbit hole on human control, agency, autonomy, delegation and the psychology behind why we trust some agents and resist others. I wrote about what I found here: The Human Need for Control in the Age of Agentic AI Let AI take over the work. Don't make it take over the meaning of the work. And if you're building this transition right now: Are we asking the right UX questions yet? Tell me what you think.
1
255
edwin emmanuel roy retweeted
Meet the new Claude mobile app! So much craft and polish across the app, including a new tab bar. Huge shoutout to @daniel__designs, @khalilcader, @pranathiperii, Andrew, and Graham for all their work across iOS and Android. Rolling out to all users, starting today!
173
159
24
4,144
224,131
Daily dose of wisdom: A "meat proxy" is a human acting as a biological relay between a generative model and another human (and their generative model), adding no cognitive value to the transaction.
26
edwin emmanuel roy retweeted
using carbon with 4 hydrogen subagents. i call it... methane.
182
659
105
11,270
231,649
Ask an AI agent the same question twice. You get two different answers. Designers can't draw a screen for every output, so stop drawing screens. Design the parts and the rules.
28
Earlier the works were burdened with the operational tasks and the manager with the power to evaluate. The question that arise with the shift is that, is the worker are capable enough to drive as a evaluator.
BREAKING: Microsoft CEO Satya Nadella declares workers will become managers of AI agents after offering 60,000 employees buyouts.
30
X is now one of the top inspiration channel for me.
23
My friends just launched Appxiom on Product Hunt today. If you build mobile or web apps with AI coding agents, Appxiom gives your agent the full bug context: what broke, where in the user journey, and how much it's hurting key goals like signup or checkout (we call it Goal Friction Impact). Works with any MCP-capable agent. They would love your support on the launch page: producthunt.com/products/app…
1
1
4
105
Huddle Global 2026 is filled with AI slop.
1
41
edwin emmanuel roy retweeted
Dario: Fuck Chinese models Sam Altman: Fuck Chinese models Jensen: Yeah, Let’s Fuck Chinese model
213
2,244
125
45,484
1,425,830
I was reading a research paper, and a something I read struck me. Even I have seen a lot of people claiming the AI generated works/codes/documents as their works as they might have spent hours crafting prompts and directing the angle. However, the generations happen in opaque black boxes with near to zero visibility for the user. Think of it like this, you are instructing someone to do certain tasks and he/she goes into another room and comes back with an answer. How can you say that this is something you created. The paper states that this as one of reasons why the US copyright law doesn’t accept AI generated content as a human authored one.
27
I was checking out the core layers of Figma right now, and to be honest, the speed is very less compared to what we have with Wonder.
29
My pressure cooker is a better AI agent than most AI agents. Put kadala on, walked away. 5 whistles = done. No spinner. No logs. Good agents don't demand attention. They whistle only when it matters.
20
Some people may even incorrectly assume that a product uses AI if the functionality matches their past experiences of AI. Your product may build on existing mental models and expectations of products that use similar technology.
1
18
I'm actually looking to set up a system using a local model that really helps me with my design. Is there any good suggestions for a local model?
23
The token limits are hitting super fast with claude. 😱😱
1
1
50
Better the CRM, lesser skilled be the salespeople.
33
I don’t like the Figma AI credits system.
23
I really liked how the @usewonder understanding the agentic experience. The ability of a user to steer an ongoing agent task is something most experiences lack. The agency this brings back to human is huge. A good UX pattern to follow. @aibek_design
1
1
50