Shashwat Jain (@JainAI29) header
Shashwat Jain (@JainAI29) profile picture

Building and sharing thoughts on AI Agents | Currently building vision AI Agents in sports world | IITM’20 https://xitter.gq/t.co/JwLCrW78qc

Delhi
Joined March 2013
Photo by Shashwat Jain (@JainAI29)Photo by Shashwat Jain (@JainAI29)Photo by Shashwat Jain (@JainAI29)Photo by Shashwat Jain (@JainAI29)Photo by Shashwat Jain (@JainAI29)
Shashwat Jain retweeted
Introducing Reality-Aware Rendering in @Simulon, a completely new way to create with the real world. Place neural assets with your phone and act out the scene. When you stop recording, Simulon renders the take, reading everything in it (what you do, what you say and the place around you), and your assets respond to all of it. The final is regenerated from your recording: a new performance of the scene, not a copy of the footage. Available now on iPhone. Coming to Android soon.
938
2,575
946
15,647
1,705,082
Shashwat Jain retweeted
We're giving away a CONTROL Resonant custom wrapped NVIDIA GeForce RTX 5080 to celebrate its official release with #RTXON. To enter: 🟢 Share this post 🟢 Comment #RTXON T&Cs: nvda.ws/4z24Gcq
37,018
30,532
2,813
33,140
3,115,100
Say hello (literally) to Gemini 3.8 Live and 3.8 Live Extended Thinking, our new SOTA live audio models, available with frontier price + performance. 3.8 Live supports 97 languages (can seamlessly switch), async tool calls, and more!
291
240
79
4,276
402,978
Shashwat Jain retweeted
The Claude Code team (@ClaudeDevs) put out a 22-minute roundtable on how they use Claude Code internally. No demos, no slides. Three engineers at a table. It's the clearest signal yet of where agentic coding is actually heading. 8 things that stuck: 1. Most of their coding happens in Slack now, not the terminal. One of them puts it at 70-80% through Claude Tag. The TUI is the "micromanagement surface" for the remaining 20%. 2. They give Claude goals, not tasks. A year ago they cared about every tool call and every line of transcript. Now they hand over an outcome and inspect the result. 3. They delete harness features as models improve. The to-do list existed because Sonnet 3.5 would do 3 of 5 tasks and quit. It's gone. Every harness feature is a patch over a model weakness, and every patch has an expiry date. 4. AskUserQuestion, a tool one of them built and fought to get Claude to call well, is already fading. He now has Claude generate an HTML artifact with mockups and diagrams that asks him the questions instead. 5. Routines run one level above the session. Daily: triage all incoming feedback, bucket by importance, fix only what Claude is highly confident about. The laptop-closing problem is what pushed them from local to hosted agents. 6. Code review inverted. The "three nitpicks to prove I read it" review is dead. Claude does the nitpicks. Humans review API structure and service boundaries, and Claude surfaces the context needed to think at that altitude. 7. Workflows came out of code review. Fan out to find bugs, run each candidate through adversarial review from 3 perspectives, coalesce. Trust comes from the orchestration being deterministic code: a for-loop won't skip an item. 8. What they miss: performance engineering and pixel-level CSS craft. Claude is better at both now. "Software engineering is the profession of change." The meta point: this team builds on a substrate that shifts every two months. Their answer isn't more scaffolding. It's less. Build a primitive, watch the model outgrow it, delete it, build the next one at a higher altitude. If your agent workflow still has you reading every tool call, you're operating where they were a year ago. youtu.be/S-sYlFiGFv8
1
115
Shashwat Jain retweeted
My Mac randomly shut down with 57 AI coding sessions running — 7 interactive Claude Code sessions and 50 subagent lanes, all killed mid-flight. Recovered every single one in under a minute. Turns out CLI coding agents (Claude Code, Codex) write every session's transcript to disk continuously — a crash loses at most the in-flight response. The hard part is finding WHICH sessions died and WHERE to resume each one from. So I built session-rescue, an open-source Agent Skill that: • correlates transcript mtimes with your machine's boot time • separates real sessions from subagent lanes and headless batch runs • prints ready-to-paste cd <dir> && claude --resume <id> commands Zero install — run it straight off GitHub after a crash (macOS/Linux/Windows, stdlib only): Built entirely with Claude Code. MIT licensed. New skills get added as they're battle-tested — PRs welcome. github.com/ShashwatJain29/cl…
1
3
148
Shashwat Jain retweeted
some tips for Claude Fable 5.1 Try Fable 5.1 at low effort | Fable 5.1 at low effort is often competitive with Opus and Sonnet on $ / task while scoring higher. on CursorBench 3.2.0, Fable 5.1 at low effort is at parity w/ Fable 5 at high effort at a third of the cost. Prompt cache reads are 4x cheaper for Fable 5.1 | cache hits dropped from $1.00 to $0.25 / MTok. if you are using prompt caching, Claude Console can monitor your prompt cache hit rate. also run “/claude-api cost-optimize” in Claude Code to run diagnostics on your prompt cache configuration. see pricing: platform.claude.com/docs/en/… Simplify your prompts | remove verification rituals, emphasis boosters, scratchpad scaffolds, stale few-shot examples, or contradictory rules. in Claude Code, run “/claude-api prompt-audit” to inspect prompts or skills for common anti-patterns. see our prompt guide: platform.claude.com/docs/en/… Change effort mid-conversation w/o breaking the prompt cache | Previously, mid-conversation changes to effort invalidated the cache. see docs: platform.claude.com/docs/en/… for migrating to Fable 5.1, see the migration guide and run "/claude-api migrate" in Claude Code to help update API configuration to use Fable 5.1. platform.claude.com/docs/en/…
75
91
37
1,408
165,094
Shashwat Jain retweeted
Today we’ve raised $52M Seed and we are announcing the public launch of S2.1 Pro. >It can clone a voice from 5 seconds of audio >2x faster than Cartesia & 1/6th the cost of Eleven Labs >most expressive model with word level control over emotion, intonation, pacing etc We support frontier AI companies including HeyGen, LiveKit, Retell, Sanas, and OpenArt all run our model in production. If you're a business and we can't cut your voice AI costs by 50%, we'll give you 1 year of Fish Audio for free. Book a demo: s.fish.audio/tmapke To celebrate our first birthday, we'll give you 1 month of S2.1 Pro for free. Like, retweet, and comment “Fish” to get it.
4,816
3,499
703
13,863
6,279,958
Shashwat Jain retweeted
Your AI agent can now generate and ship videos. HeyGen CLI is now live. Run one command and your agent handles it all: script → avatar creation → video → delivery All from the terminal. Just your agent and the CLI. RT + Comment “CLI” and we’ll DM API credits (must follow)
1,052
623
124
1,533
226,821
1/ Gemini 3 in action, a quick🧵  You can give Gemini 3 anything (images, pdfs, scribbles, etc) and it will create whatever you like: an image becomes a board game, a napkin sketch transformed into a full website, a diagram could turn into an interactive lesson.
236
761
143
6,702
748,748
Shashwat Jain retweeted
Announcing Grok CLI 🔥 An open-source AI agent that brings the power of @grok directly into your terminal. Built over the weekend with a few design principles in mind: ✅ No LLM frameworks ✅ No lobotomizing models ✅ Hackable (MIT licensed) Github link below.
316
702
173
5,719
1,007,737
This AI Agent replaces your $200K/Year Marketing team while I was eating shawarma at 2am, it scraped 6 platforms, analyzed 73 comment threads, and handed me 9 deploy-ready scripts. It stalks your niche, steals what’s working, and turns it into viral content built to print attention. here’s what the system does: – scrapes TikTok, IG, LinkedIn, YouTube, FB, and Twitter – finds breakout trends, emotional triggers, and dopamine hooks – maps audience avatars + repurposing angles – generates platform-native scripts with hooks, CTAs, SEO intros – formats everything into a deploy-ready content playbook no more brainstorming. no more “what should we post today?” it’s like giving your intern the mind of Alex Hormozi - on a 400mg caffeine drip. the workflow includes: - Multi-platform scraper engine - Viral comment sentiment analyzer - GPT-powered creative generator - Smart repurposing module - Auto-formatted Google Docs output this isn’t ChatGPT with a cute prompt. it’s a weaponized idea machine. if you’re serious about content — and tired of guessing — this flips the game. Comment “ENGINE” + repost this + follow me I'll DM you everything in the next hour skip this, and go back to scheduling posts that get 4 likes.
1,805
986
29
2,384
341,703