Asad Ali

Building agentic harness that allows chatting over Microsoft Graph API

@adxtyahq deferred

day 1 of learning golang gonna document the entire journey through daily timelapses

@KaiXCreator all these things makes you founder x.com/asad_ch3/statu…

If Claude wrote almost all your code, what actually makes you the founder?

AI makes prototypes cheap. The engineering ladder now looks different: Can you make it reliable? Can you test it? Can you debug it? Can you secure it? Can you explain it? Can others maintain it? Can users trust it? The real engineer shows up after “it works once.”

@thinklikekai @dr_cintas mf, you spoke that shit

@asad_ch3 @dr_cintas you bypass it dawg

OpenClaw just got an unfair advantage over every other AI agent 🤯 It can now use Scrapling to scrape any website without getting blocked by Cloudflare. You don't need to maintain selectors when websites update their structure. - 774x faster than BeautifulSoup. - Zero bot detection. - Bypasses ALL Cloudflare protections natively 100% Open Source.

Sub-agents are a promising inference-time scaling primitive: • Expand an agent's working memory • Divide-and-conquer hard problems • Solve problems faster with parallel execution But how do we train a model to best take advantage of sub-agents and make sure we get these benefits? Very excited to release RAO: Recursive Agent Optimization. RAO is an end-to-end reinforcement learning approach for training LLM agents to spawn, delegate to, and coordinate with recursive copies of themselves (that can themselves spawn other agents) - turning recursive inference into a learned capability. 1/10

@adrian_horning_ Agents can still not fix deep technical problems. Also this will make the application vulnerable to attackers in my opinion.

Someone should productize this: > Email comes into support > Agent responds > Agent fixes problem, creates PR > Another agent reviews the PR, makes sure tests pass and merge > Tells customer its live People could OAuth with their google accounts Million dollar business

@micheltamanda SGLang is better alternative to vLLM

My AI stack: 4x RTX3090, 1x 5060, 1x 4070, 1x M3 Ultra. VRAM 112GB, Unified Mem 256GB Backend: vLLM, Ollama Models: qwen3.6 family, whisper What would you do next?

@Taniyatweets_ Even 6 year m1 air is still so much good

Why do so many developers prefer MacBooks?

Quail in the hands of blind 🤣

@CodeWithAmann Switched back to Vs code from cursor

Be honest, which IDE are you opening first every day?

Open-weight models still seem a bit behind frontier models in reliable tool calling and structured outputs. I still run into malformed JSON, inconsistent function calls, and agents losing track in multi-step workflows more often than expected.

@thdxr Doesn't LiteLLM do the same?

we're working on a library to abstract over all the llm providers there's very few teams that have dealt with the quirks between providers at the scale we have it's written in effect but will also have a vanilla api progress is in the opencode repo under packages/llm

@HelloJarbis @dr_cintas not exactly, does it works on CF turnstile?

@dr_cintas Cloudflare protection was the silent killer of my early scraping projects. This seems genuinely useful — native bypass without the arms race.

@thinklikekai @dr_cintas Wtf you saying, bypass cloudflare is not piece of cake

@dr_cintas 774x faster than BeautifulSoup Bypasses Cloudflare Zero bot detection This is the kind of tool that makes entire scraping companies obsolete overnight The moat in web scraping just disappeared

@caspar_br @BraceSproul I am not sure about terminal safety on open weight models, just single destructive execution can ruin the project, so i think for that it needs to be actively supervised and should not be running autonomously.

.@BraceSproul changed our org's internal model in Fleet from Sonnet 4.6 to Kimi K2.6 and I didn't even notice. Open models are already good enough for most tasks, though not the hardest coding work yet, and are 5-10x cheaper... 2026 is going to be fun

@DavidOndrej1 i am a lite user so sonnet 4.6 works for me on pro plan

Opus 4.7 is unusable today

@PhiloGroves @atKensai @claudeai Yes, you raised a good point but this may be really helpful for the people who want to work part time quick usage or the one on claude max plan

@atKensai @claudeai Now you need to pace yourself more, unless you want your weekly limit to be gone twice as fast. I was already hitting the weekly limit on day 3 before I cancelled, this would do nothing for me.

We’ve agreed to a partnership with @SpaceX that will substantially increase our compute capacity. This, along with our other recent compute deals, means that we’ve been able to increase our usage limits for Claude Code and the Claude API.

@DavidOndrej1 What is the best open harness for coding?

Cursor Agent might be a better harness for Opus than even Claude Code itself

@trikcode Just started the journey. How long it should take for me to get a remote opportunity?

Your network is your biggest advantage in tech. Follow other builders. Reply. Engage. That’s how opportunities start.