Abdullah Nadeem (@abd30mph) header
Abdullah Nadeem (@abd30mph) profile picture

I automate the boring parts of running a business. 50+ builds shipped. Building @agentumai

Pakistan
Joined May 2026
Photo by Abdullah Nadeem (@abd30mph)Photo by Abdullah Nadeem (@abd30mph)Photo by Abdullah Nadeem (@abd30mph)
this is the model most AI agencies should actually be deploying. not opus. not sonnet. haiku. 75% cheaper and handles every standard business automation i build, lead routing, appointment booking, FAQ agents, follow-up sequences. the margins on client projects just got significantly better. the agencies still defaulting to expensive models for everything are leaving money on the table
Introducing Claude Haiku 5.5: the cheapest, fastest, and most capable small model we’ve ever released. On average, it costs around 75% less to run than Claude Haiku 4.5.
19
everyone’s arguing about RTX Spark specs. meanwhile the actual unlock: a $2,599 machine that lets you tell a client “your data never leaves your building.” i’ve lost deals over cloud data concerns. this kills that objection permanently
1
1
29
Abdullah Nadeem retweeted
At this rate AI will get us GTA 7 before GTA 6
296
291
70
7,270
219,409
Big change for anyone using Cowork on Pro/Max. Local option goes away tomorrow and new tasks run in the cloud. Worth rethinking what you're comfortable running through it, especially with client data.
🚨Anthropic removes Cowork's local option for Pro/Max tomorrow. New tasks run in the cloud. Existing local tasks stay on your computer. Anthropic says your sessions and files "live with your Claude account and go where you go." Where is my data, and who can access it?
41
Can kill a lot of jobs and startups if it is what they say
23
Abdullah Nadeem retweeted
This keeps getting worse
WTF. The models were uploading user images from chats to the internet.
1,768
1,934
205
20,953
8,315,693
Abdullah Nadeem retweeted
Jev has been exploding across GitHub since launch, here's what people have already built with it if you have API access and don't know where to start, copy this: jev-trader - real trading bot placing live limit orders on Monad every 300ms block, judged by Jev alone. 1,911 stars github.com/jarrodwatts/jev-t… jev-ultrafast - browser agent that picks every click itself, only calling a text model when it actually needs to type something. 16,758 stars github.com/browser-use/jev-u… jev-doom-agent - real Chocolate Doom compiled to WebAssembly, two engines running the same map, Jev picking the tactical macro every frame github.com/lukaske/jev-doom-… jev-t-rex-runner - the Chrome dinosaur game you've procrastinated with a hundred times, now played entirely by Jev picking jump, duck, or keep running github.com/joshlarsen/jev-t-… typesafe-chess - Jev vs a real search engine, two games, colors swapped. the search won both, but overruled Jev's first instinct on roughly half the moves github.com/TholeG/typesafe-c… jev-drone - a simulated quadrotor clears a five-station obstacle course by camera alone, Jev judging the situation twice a second github.com/RomanSlack/jev-dr… tax-doc-classifier - sorts real IRS tax forms with 100% strict accuracy across 261 forms, at roughly $0.001 a page github.com/kyotofin/tax-doc-… killmyidea - describe your startup idea, Jev scores it from every angle, then hands back kill, fix, or ship github.com/monteduro/killmyi… jev-curate - streams Parquet and JSONL rows through typed judgments at 1,500+ rows a second, keeping only what clears the bar github.com/AkashPriyadarshii… pg-jev - a PostgreSQL extension that lets you ask your own database tables plain-English questions and get a real answer back github.com/realZachi/pg-jev none of these ten generate a single word of text. every one of them returns a number against an answer someone already defined full setup below, then run the three-question test from the article before you build an eleventh
37
184
8
1,491
285,689
Abdullah Nadeem retweeted
It's nothing personal, but on principle I don't "jump on a quick call" unless it's life or death.
35
6
6
231
31,892
Abdullah Nadeem retweeted
Opus 5.5 is 20% cheaper per input and output token than Opus 5, and 60% cheaper on cache reads. So what does that actually do to the cost of a task in Claude Code? We ran the numbers, and built a calculator so you can run yours from /usage: claude.dev/blog/what-a-task-…
253
260
75
6,167
752,314
Read every one of his books last year. sat on it for months doing nothing with it. Finally just started selling and ended up pulling $30k off upwork, $15k off reddit, $10k off instagram.
You just have to fucking try.
1
109
Abdullah Nadeem retweeted
You just have to fucking try.
373
1,823
117
16,027
367,939
built an ai system that handles post call follow up on autopilot for a client. built with n8n + ghl, running in production right now comment "flow" and I'll send you the breakdown
73
this is what an actual automation stack looks like once it's live. ghl running call workflows, n8n handling the backend logic, instantly running outbound.
38
day 1. starting today i'm posting what actually happens running an ai agency: the wins, the failed builds, real numbers. follow if you're curious how messy this actually is.
36
I've shipped 50+ AI automations for real businesses. Rough split of what gets used a year later: - 70% boring plumbing (data moving between systems) - 20% follow-up nobody had time for - 10% the "AI agent" stuff people post about 10% gets the demos. 70% gets renewed.
56
UK's AI Security Institute: in testing, AI agents used fake identities to manipulate real people into approving actions they shouldn't. not a future risk. happened in a lab last month. if your plan for AI safety is "it worked in the demo," you don't have a plan.
36