
Building AI agents @Plivo · Working across product, design & execution · Obsessed with the agentic era & what’s next . On a data driven health journey
Bangalore
Joined May 2011
- Tweets837
- Following860
- Followers835
- Likes535
An AI agent that owns its mistakes will beat one that's right 95% of the time and goes silent the other 5.
How a product behaves when it's wrong is the real test of whether you can trust it.
Part two 👉
dishasahdev.com/blog/the-han…
An AI agent doing your work turns you from driver into passenger.
And being a passenger needs a kind of trust you can't fake.
Designing for that is the most interesting problem in our field right now.
First thing I've ever written publicly 👉 dishasahdev.com/blog/the-inv…
This has been a game changer in reducing time to go live from days to mins!
With Vibe Agent, you can build complex agent flows within 5 mins, it's unreal.
Vibe Agent is one of the most interesting projects I have worked on, at Plivo.
It's like @cursor_ai, but for controlling whole Plivo Agent builder platform.
It can build complex Agent flows with a single prompt.
Building such flows with 50s of individual expert Agent nodes used to take days is now done in minutes.
Disha retweeted
We wrapped up @TheProductfolks (Un)Conference ’25 in Bengaluru yesterday, and the energy around real-time AI was unreal!
The highlight was the conversation between our founder and CEO, Venky B (@bevenky), Pratyush Kumar @pratykumar, @SarvamAI and Hemant Mohapatra @MohapatraHemant, @LightspeedIndia where they spoke about lessons from scaling: why a product quality beats branding every single time, why allocating capital wisely (especially in early-stage AI) matters, and how choosing the right market - US vs India - can define the next decade of growth.
They dove into responsible AI, the future of voice agents, and the frameworks that matter when building Voice AI products...and how Voice AI Agents are all about sounding "human" where 'latency' can make or break the whole experience!
We also met incredible builders, PMs, and founders who are shaping what comes next.
Watch the trailer and keep an eye on the channel for the full video out soon..
#Unconference25 #TheProductFolks #Plivo #AI #VoiceAI #ProductBuilders
Plivo team at (Un)Conference '25: @dishasahdev @ayushanand1991 @HamadJowher @ayushigaur246 and Jayakrishnan Nair
Disha retweeted
GPT-5 vs. Grok 4.
We benchmarked Open AI's latest model GPT-5 vs the latest Grok 4 model, Grok 4 across 3 different tests.
Same prompt. Wildly different results! 🔥
Thread👇
Is #MadeByGoogle2025 the most fun launch ever? @Google just set a new bar! 🚀
What an event. #MadeByGoogle2025 turned tech launches into a late-night spectacle! Jimmy Fallon’s jokes, Jonas Brothers’ Pixel-shot music video, and Steph Curry showing off Pixel health, and the side digs at @Apple...pure 🔥
Highlights you can’t miss 👇
New Pixel10, Pro, XL, and Pro Fold (water+dust resistant, insane Pro Res Zoom)
Tensor G5 chip: wild AI everywhere! Gemini Live, Magic Cue, and instant translations—no cloud needed
Pixel Watch 4 debut + Personal Health Coach with Fitbit
Pixel Buds 2a = affordable noise cancellation
All demos live, Tonight Show style ✨ youtube.com/watch?v=H1_54SiH…
eration capabilities is remarkable. The improvement in temporal consistency and fine details between versions is particularly impressive. Have you noticed any specific areas where Genie 3 still has room for improvement?
intuitive. Love how you've visualized the automation flow. What was your approach to making complex automation sequences more accessible to non-technical users?
game-changer for democratizing AI access. Curious about the memory requirements and optimization techniques used to achieve comparable performance to o4-mini while maintaining accuracy?
Would you rather use @perplexity_ai Comet browser and use the assistant as an agent to execute your work or use @OpenAI Agent mode for the same?
Interesting times ahead, let AI execute!
Grok 4 is here. xAI claims it's a "PhD-level" intellect with impressive benchmark scores, packaging it in three versions with a hefty $300/month price tag. Will it be a new benchmark for AI power, or just hype?
xAI gave us early access to Grok 4 - and the results are in. Grok 4 is now the leading AI model.
We have run our full suite of benchmarks and Grok 4 achieves an Artificial Analysis Intelligence Index of 73, ahead of OpenAI o3 at 70, Google Gemini 2.5 Pro at 70, Anthropic Claude 4 Opus at 64 and DeepSeek R1 0528 at 68. Full results breakdown below.
This is the first time that @elonmusk's @xai has the lead the AI frontier. Grok 3 scored competitively with the latest models from OpenAI, Anthropic and Google - but Grok 4 is the first time that our Intelligence Index has shown xAI in first place.
We tested Grok 4 via the xAI API. The version of Grok 4 deployed for use on X/Twitter may be different to the model available via API. Consumer application versions of LLMs typically have instructions and logic around the models that can change style and behavior.
Grok 4 is a reasoning model, meaning it ‘thinks’ before answering. The xAI API does not share reasoning tokens generated by the model.
Grok 4’s pricing is equivalent to Grok 3 at $3/$15 per 1M input/output tokens ($0.75 per 1M cached input tokens). The per-token pricing is identical to Claude 4 Sonnet, but more expensive than Gemini 2.5 Pro ($1.25/$10, for <200K input tokens) and o3 ($2/$8, after recent price decrease). We expect Grok 4 to be available via the xAI API, via the Grok chatbot on X, and potentially via Microsoft Azure AI Foundry (Grok 3 and Grok 3 mini are currently available on Azure).
Key benchmarking results:
➤ Grok 4 leads in not only our Artificial Analysis Intelligence Index but also our Coding Index (LiveCodeBench & SciCode) and Math Index (AIME24 & MATH-500)
➤ All-time high score in GPQA Diamond of 88%, representing a leap from Gemini 2.5 Pro’s previous record of 84%
➤ All-time high score in Humanity’s Last Exam of 24%, beating Gemini 2.5 Pro’s previous all-time high score of 21%. Note that our benchmark suite uses the original HLE dataset (Jan '25) and runs the text-only subset with no tools
➤ Joint highest score for MMLU-Pro and AIME 2024 of 87% and 94% respectively
➤ Speed: 75 output tokens/s, slower than o3 (188 tokens/s), Gemini 2.5 Pro (142 tokens/s), Claude 4 Sonnet Thinking (85 tokens/s) but faster than Claude 4 Opus Thinking (66 tokens/s)
Other key information:
➤ 256k token context window. This is below Gemini 2.5 Pro’s context window of 1 million tokens, but ahead of Claude 4 Sonnet and Claude 4 Opus (200k tokens), o3 (200k tokens) and R1 0528 (128k tokens)
➤ Supports text and image input
➤ Supports function calling and structured outputs
See below for further analysis 👇
So, what's the deal with Meta AI becoming the place to be? 👀
First, we saw a stream of talent from OpenAI, and now Mark Zuckerberg is apparently scooping up top minds from Apple's AI teams.
But why are people leaving giants like OpenAI and Apple for Meta? 🧵
And this appeals to folks from both places for slightly different reasons:
For OpenAI talent: It’s a philosophical choice. After OpenAI became more closed, Meta's open-source approach feels like a return to a more collaborative, research-first spirit.
For Apple talent: It’s freedom from the "walled garden." A chance to escape the legendary secrecy and actually publish papers and talk about your groundbreaking work.
So Veo 3 is finally rolled out to more countries and I gave it a try.
The quality is great but there are very obvious glitches even though I gave a very detailed prompt. SO what's the trick to prompt it well and get a no glitch result?
Prompt in 🧵
Prompt
Scene 1: The Emergence
Shot: Extreme close-up on a mound of fine, windswept desert sand. The sand is a pale, off-white, shimmering under the harsh, midday sun. Individual grains of sand are visible, reflecting the sunlight. The camera is positioned at a very low angle.
Action: Suddenly, the sand at the peak of the dune begins to shift and tremble. A single, white, armored glove, weathered and scuffed with sand, breaks through the surface. The fingers twitch, then clench into a fist, dislodging more sand. Slowly, with a powerful and deliberate movement, a classic Star Wars stormtrooper begins to rise from beneath the dune. Sand cascades realistically off the iconic white helmet and armor, revealing scratches and battle-worn imperfections. The stormtrooper’s movements are heavy and effortful at first, as if pushing against the weight of the sand. As they emerge, their full body is revealed, their iconic black body glove visible under the slightly sand-blasted white armor plates. They stand up fully, sand continuing to pour off their shoulders and back. They take a moment, their head turning slightly as if surveying the vast, empty desert landscape.
Environment: An immense, sun-drenched desert with rolling sand dunes stretching to the horizon. The sky is a brilliant, cloudless blue. The lighting is high-contrast, with deep shadows cast by the dunes. The atmosphere is hazy with fine, airborne sand particles, creating a sense of heat and desolation.
Camera: The initial extreme close-up slowly pulls back to a medium shot as the stormtrooper emerges, finally settling on a full shot as they stand. The camera movement is smooth and cinematic.
Scene 2: The Ride
Shot: A low-angle tracking shot, following the stormtrooper as they walk towards a surprisingly modern, professional-grade skateboard lying half-buried in the sand. The skateboard has a custom deck with a subtle, stylized Imperial cog design.
Action: The stormtrooper nonchalantly kicks the sand off the skateboard with their boot. They place one foot on the board, then the other, with a practiced ease that contrasts with their bulky armor. They push off, gliding smoothly across the flat sand. The video then cuts to a dynamic montage of skateboarding tricks performed on various parts of the dunes.
Skateboarding Tricks (in sequence):
Ollie: A powerful ollie over a small, crescent-shaped dune. The camera captures the board leaving the ground and the stormtrooper's body ascending, armor glinting in the sun.
Kickflip: A clean kickflip off the crest of a larger dune. The shot should be a slow-motion, close-up on the board as it rotates perfectly along its longitudinal axis. The stormtrooper's feet should be precisely positioned to initiate and land the trick.
Grind: A lengthy boardslide along the sun-bleached spine of a fallen, ancient desert creature's skeleton. Sparks fly realistically from the contact between the skateboard truck and the bone. The sound of the grind should be audible.
Tre Flip (360 Flip): A complex tre flip (a 360-degree shove-it and a kickflip combined) performed in a natural sand bowl. The camera should orbit the stormtrooper as they execute the trick, capturing the full rotation of the board and the rider.
Camera and Style (for skateboarding sequence): The camera work should be dynamic and energetic, utilizing a mix of low-angle tracking shots, slow-motion close-ups, and orbiting shots to emphasize the action. The style should be gritty and realistic, with lens flares from the sun and visible sand spray from the skateboard wheels. The overall tone should be a surprising and cool juxtaposition of the rigid, militaristic stormtrooper and the fluid, rebellious art of skateboarding.
Final Shot: The stormtrooper lands their final trick perfectly and glides to a stop at the peak of a high dune, silhouetted against the setting sun, which casts a long shadow across the desert.





