Building with AI coding agents and posting what actually works. AI @AudaciousHQ. Ex-CTO @GoCodeoAI. IIT Kanpur

SF
Joined October 2022
every new session i re-explained the same port, the same constraint, the same fix we already ruled out. the agents were never the bottleneck. i was the memory between them. so i built coding brain.
21
1
1
39
12,646
openai tested moving old cobol programs to java and python, where the output has to match byte for byte. gpt-6.1 sol got 86.6 percent right, the previous model got 5.4. i'd still replay a year of real transactions through it before switching anything.
1
87
claude code was building a temperature monitor when it noticed the readings were too high. it checked the cpu and found a cryptominer running at 1175 percent. i think agents that watch their environment will catch a lot of problems humans miss. xitter.cf/IntCyberDigest/status/…
❗️ A Mac user says Claude Code was building them a temperature monitor app when it found a cryptominer using 12 cores. The readings looked too high, so it checked the CPU and found XMRig at ~1175%, with the chip at 112°C.
97
cloudflare open sourced the 450 line skill behind its own vulnerability harness. the rule that makes it usable is that the agent checking a finding is never the agent that found it. 20,799 raw candidates became 7,245 real bugs.
1
68
companies post ai engineer jobs but they need data engineers who know kafka and spark. rag tutorials skip the pipeline work. i think the ai boom made everyone forget you need infrastructure to move data before you can prompt it. xitter.cf/swyx/status/2112073533…
Data engineers > AI engineers there's a big difference between a "RAG guy" and a "Data Intensive Applications" person like in this screenshot and there's never been a starker difference in demand in my entire career in tech (aka post search engine wars) if your blog / github is full of stuff like this, I have a fantastic mission/job for you (SF preferred, can relocate u, visas ok) (if you can name who's blog this is without even checking, you definitely are in the community. bonus points if you know where he ended up also without looking) [email protected] and the team will reach out if you fit
1
104
delta amazon chase and yelp started blocking ai agents. agents skip the ads and upsells so companies detect them and shut them out. i think this ends up like ad blockers, the detection gets better and the evasion gets better and nobody wins.
1
1
62
ai made coding games more fun than playing them. people build little game pieces they never finish instead of loading up the games they bought. i think this is what happens when tools get better than the product. xitter.cf/trq212/status/21119453…
I've realized I often would prefer to vibe code little parts of games than actually play games these days. Especially cerebral ones that require some thinking like League. I'm not quite sure what this really means for games but I think it's important?
56
i don't trust a chart i can't check. so every chart in localdash has a SQL button. csv in, live dashboard out, runs on your machine. (1x speed btw) github.com/jatingargiitk/loc…
50
engineers using ai to write code and ai to review the same code. both sides run the same model so the review adds zero new signal. the bottleneck moved from writing to deciding what actually matters. xitter.cf/Yuchenj_UW/status/2111…
90% of software engineering in 2026: Engineer 1 → AI: “Build this and open a PR” Engineer 1 → Engineer 2: “Can you review?” Engineer 2 → AI: “Review this PR” Engineer 2 → Engineer 1: “LGTM” The irony: both engineers use the exact same AI model.
1
60
google open sourced harness2. an agent proposes edits to its own prompt, tools and memory, runs the same task under each, and keeps only what the runs back up. 11.5 points better with the weights frozen. the scaffolding was the undertrained part.
87
odyssey put a fruit fly brain inside their world model. the fly navigates completely ai generated worlds on its own. simulating real biology in ai worlds is more interesting than just physics engines.
100
openai added three new models with color names: aurora-red-alpha, aurora-green-alpha, aurora-blue-alpha. first time they broke from gpt numbering. i think this is a new tier or capability that needs its own brand instead of another version. xitter.cf/AndrewCurran_/status/2…
Three mysterious new models just showed up in the OpenAI API: aurora-red-alpha aurora-green-alpha aurora-blue-alpha
1
144
glimpse is a visual model that understands video and changing angles, not just single images. team is ex-gemini, apple and xai leads. i think making visual ai work on movement is more useful than getting 2 percent better on static benchmarks.
1
67
deepseek v4.1 scores 77 on coding benchmarks, matching opus 5 performance at 83 times lower cost. the open source model caught up to the best proprietary one in under a week. i think the cost gap alone makes this the new baseline for most coding. xitter.cf/Yuchenj_UW/status/2111…
The "OSS will never catch up" takes didn't survive a week. DeepSeek-V4.1-Flash-1017 scores 77 on DeepSWE v1.1. Opus 5-level performance at 83x lower cost. Incredible for a model of this size. We will be launching it on Databricks! Never bet against open source.
138
the nist curve seeds are lrand48 output seeded with a unix clock from august 1997. lrand48 hides most of its state, so claude recovered it with lll lattice reduction and hit p-256 at candidate 1,842. a recipe you can rerun beats a story nobody can check.
54
ai turns a text prompt into a live dashboard with charts and real data. you describe what you want and it builds the whole thing. i think data analysis just became accessible to everyone.
63
claude code can hack smart tvs to remove ads, bloatware and tracking. you send it the tv listing, ask can you hack this, and it reverse engineers the firmware. i think this is what ai coding was made for, projects nobody would pay someone to do. xitter.cf/doooyle/status/2111196…
Hacked my TV with Claude Code today No more ads, bloatware, spyware or laggy UI I also got a beautiful screensaver with my revenue, calendar, weather, and notifications for new Stripe payments It was super easy! I sent it the listing for my tv, asked "can you hack this?" And it talked me through how
282
cursor agents can build interactive canvases while they work. you ask for a dashboard or report, the agent makes it live, and you get a shareable link when it's done. i think this is how all agent outputs should work, not just text logs. xitter.cf/cursor_ai/status/21111…
Cloud agents can now build interactive canvases. Ask for a dashboard or report while the agent works, then share the finished canvas with your team as a link.
2
112
claude solved a 25 year cryptography mystery. the nsa gave the internet curve seeds for ecdsa in 1997 and nobody knew where they came from. claude figured it out in 30 minutes. i think this is the first time ai answered a real open security question.
1
1
1
168
nvidia app added an mcp server that lets ai agents control recordings, screenshots and instant replay. runs locally, enabled by default. i think this is the first gpu vendor shipping agent hooks in consumer software. xitter.cf/Pirat_Nation/status/21…
The latest NVIDIA App update (version 11.0.9.565) adds an AI agent MCP server, enabled by default, runs locally and lets AI tools interact with the NVIDIA App. It can: • Check the current game • Start/stop recordings • Take screenshots • Control Instant Replay and Highlights • Toggle RTX Dynamic Vibrance • Show performance stats • Find and launch games • Check NVIDIA drivers • Change game optimization settings • Adjust Battery Boost and Whisper Mode on laptops Turn off that shit
1
1
114
openai retired atlas, its own browser, in july. the chrome extension that replaced it now reads your open tabs and acts in them, and it has 7 million users. a browser agent only works if it inherits the session you are already signed into.
2
2
67