Gaurav Devsarmah

[Gaurav Devsarmah @g_devsarmah]

One of the challenges with new tech is that there seems to be a period of creative exploration and building of things that are very cool but are detached from monetary value - believe this is why we are seeing some amazing things with AI (non model level) right now but commercial value for a lot of it is 0

[Gaurav Devsarmah @g_devsarmah]

Incredible that this was laid out in 2001 by @erichorvitz , the tech is finally scalable now and personal agents are running hot but we still haven't nailed the right human-AI interface

https://t.co/g0sIOPQcEP

[Gaurav Devsarmah @g_devsarmah]

I wanted to play around with @OpenAI 's (@OpenAIDevs) new Agents API and Realtime live voice so decided to build a little Lunar city simulation and an AI co-pilot 'Pip'.

You can talk to Pip while work happens around you: ask for more solar power, send a robot to build an array, or point at a habitat and ask about it. The conversation continues while the work runs in the background.

Wanted a fun way to explore what happens when a voice interface is connected to agents that can take action and a coding harness that can generate and test missing capabilities within defined guardrails.

I'm sure we are going to see more and more experiences where the software adapts to the user rather than the other way round.

[Gaurav Devsarmah @g_devsarmah]

@levie we will need to build systems to keep reducing volume. right now the focus is on the generation/additive part (more code, bigger shortlists in recruitment pipelines, more info about your prospects etc.) and not enough on better curation

[Gaurav Devsarmah @g_devsarmah]

AI powered work has such a different loop than a traditional approach. you constantly have to make sure you are limiting between 3-4 options of anything (designs, prototypes etc.) so you pick a direction (or not) out of a limited set rather than 20 finished versions before we’ve agreed on whether any of it is going in the right direction. Agentic systems and human evaluations need to be built around constant judgements to keep reducing the options in the outputs

Judgment needs to move way earlier in the work.

[Gaurav Devsarmah @g_devsarmah]

codex computer use is so under rated and barely utilised outside of the tech world

[Mitchell Hashimoto @mitchellh]

I can’t stress enough how little an idea matters compared to the agency of the people executing the idea.

I have had the privilege of knowing and sometimes even working with some of the most successful people (by various metrics).

The difference between mediocre and excellent work and outcomes is predominantly one of agency.

In practice this means: they dont wait for things to happen to them they go out and make things happen for them.

They don’t wait for someone else to do something, for someone to teach them, for someone to give them the path, etc. They just go out and find a way to do it.

I think the single biggest superpower these people have is the realization/belief that the world around them is completely mutable. Most everything that happens is because a person made it happen.

I used to tell people to look around the room you’re sitting in. Look at everything. Every noun. It almost all exists because a person willed it into existence. Nothing is stopping you from doing the same.

I see people online all the time dismissing someone else’s success because “I had that idea first” or whatever. I mean… yeah? If so then the difference is… you. So a bit of a self own whenever I hear that.

Number one tip: act with agency.

[Gaurav Devsarmah @g_devsarmah]

Not sure about the death of SaaS but there is certainly SaaS fatigue in enterprises.

Over the last 2 decades we have gotten used to dealing with tons of different SaaS products, which at their core are basically a business relevant UI wrapped on top of databases. This worked well when software was mainly used as a system of record but as we move towards software doing things (system of action) and not just storing things, the way we design interfaces and user experiences needs to change quite a bit.

AI native players have a big advantage in redesigning the ways an enterprise interacts with new systems and solutions (without the old baggage), for example when AI does the work, you need the UI/UX to be transparently showing what's happening, controls for humans to steer at any point, newer ways to multi task within the same platform etc. This is something that legacy systems have found and will keep finding it hard to adapt to as not many firms will be ready to completely change the way their very suuccessful software has worked in the last 20 years.

This doesn't mean enterprises will completely change the way they work overnight, but it does mean with the right balance of incentives for enterprises to actually change their established ways, we have an exciting time to redesign the way knowledge work gets done.

[Gaurav Devsarmah @g_devsarmah]

Private evals and specialised org level benchmarks are a whole new level of infra that will need to be built. Most organisations right now don’t have clear ‘what good looks like’ frameworks. Designing for this from day 1 of an enterprise deployment has never been more important

[Satya Nadella @satyanadella]

https://t.co/vLmiBKTtX3

[Gaurav Devsarmah @g_devsarmah]

structured traces and evals!

[Greg Brockman @gdb]

OpenAI for self-improving tax agents:

[Gaurav Devsarmah @g_devsarmah]

so much exciting work to do in adaptive and intent based interfaces/UX

[Google DeepMind @GoogleDeepMind]

We’re reimagining a 50-year-old interface - the mouse pointer - with AI. 🖱️

These experimental demos show how people can intuitively direct Gemini on their screens using motion, speech, and natural shorthand to get things done 🧵

[Gaurav Devsarmah @g_devsarmah]

ss someone deep into building generative UIs and conversational agentic interfaces, this is a super cool development in creating more engaging and collaborative human-ai interfaces, opens up so many possibilities. 2 of my pressing issues with the current conversational and interactive AI experiences are that 1) the AI element can mostly do one thing at a time during a conversation (either listen or speak or do something but not all 3 at once) and 2) it doesn't jump in and interrupt me unless i stop talking etc., but when such models can actually proactively jump in to do things, guide me, correct me, react to what they see, so many previous experiences that felt robotic become way better.

also today's AI powered user experiences lack the multiplayer collaborative dimension (think Google Docs or Figma) but this provides a glimpse of some very cool things to build in diverse arenas from CX to robotics!

[Thinking Machines @thinkymachines]

People talk, listen, watch, think, and collaborate at the same time, in real time. We've designed an AI that works with people the same way.

We share our approach, early results, and a quick look at our model in action.

https://t.co/AFJZ5kH7Ku

[Gaurav Devsarmah @g_devsarmah]

This isn't about how AI is doing everything for me but more about the sheer difference a couple of months can make in this space to not just create efficiencies through automation but change the way we work, interact with clients and build unique experiences for clients by providing them with a ton of value long before they have even interacted with the product itself.
With the increased capabilities of AI tools, one of my key learnings off late has been that every interaction be it internal tools or external client facing non product related tools can now be made into an experience in itself.

The wow factor does not need to be just at the product level, it can be built into every aspect of the discovery, and sales processes, which creates an entirely new experience than interactions just based on word docs or excel sheets.

I'd say historically the elements between value creation, product experience and sales processes were siloed whereas now we can merge all of them to crate unique experiences at every level, which makes for a way better (and more fun) experience for all parties involve

[Gaurav Devsarmah @g_devsarmah]

@mernit anyone thats tried building even a simple chatbot on a small firm's data/context, knows how true this is. absolute nightmare when theres no naming conventions, misplaced docs etc. so with models becoming way more capable just giving them the entire thing solves so many issues

[Gaurav Devsarmah @g_devsarmah]

@DavidOndrej1 always fascinated by the outcome based pricing because in theory this is where we are heading towards but there's pretty much no work being done to build the infra needed to get outcome based pricing work especially in an enterprise setting.

[Gaurav Devsarmah @g_devsarmah]

the more powerful AI tools get, the more it becomes a question of how do we go from a culture of being trained at every level to optimise within our constraints to a culture of generating our own direction when anything is possible

[Gaurav Devsarmah @g_devsarmah]

One thing I’ve experienced again and again in GenAI: the speed of learning is the real breakthrough.

The sheer speed and efficiency at which I can test out multiple hypothesis', features and solutions on a smaller scale to have an intuition on what works, what doesnt work and how we can get to the ideal state has really made a big difference to the way I work and also to the things I can commit to.

Previously it would have taken months and months of back and forth over emails and sharing of documents to get to the point of having access to this info and also more stakeholders would have been required to be involved.

We can now utilise all these hyper powerful tools to get to the place where an organisation has a solid amount of data and intuition on what needs to be done to get to their ideal state so much quicker and in most use cases with a lot more more valuable data points.

[Gaurav Devsarmah @g_devsarmah]

There’s such a big gap between what an AI model can do in a lab and in casual settings v/s what it can do for an enterprise level solution

[Gaurav Devsarmah @g_devsarmah]

Who wouldve guessed one fundamental change from being a mad user of AI tools is you start being more patient with humans and add as much context as needed without large scale assumptions of what the other party knows

[Gaurav Devsarmah @g_devsarmah]

@signulll wouldn't you say you can design products that have 2 use cases where certain features are optimised for their use within chatgpt to get people a taste of your product and then actual valuable features/context is all in your platform