Ahmed Janim

@jackprice That’s true, but vide coded websites don’t look crappy , they look the same but not bad, I would argue that vibe coded websites perform better than handwritten ones

@janim007 Because people don’t trust crappy vibe coded websites. So the look and feel absolutely matters. You’re not going to go to a doctors office that looks like it just got burned down 😭

90% of website now have the exact same UI Do you think this is an issue?

@daicandev Ones that don't forget. Most AI tools today have amnesia — every conversation starts from zero. The products that persist context across time and surfaces will win.

Builders and founders, If everyone can build with AI now... What will actually make a product succeed?

@T_Zahil Each one lives in a different surface — terminal, browser, desktop. The real win isn't which agent you pick but when they share context across surfaces so you're not the human clipboard between them.

Please someone explain to me why should I use Hermes if I already use Codex, Claude etc What could I do with it?

@diegocabezas01 Platforms absorbing their own agent startups is the predictable play. The interesting question is what happens when users want agents that span across platforms — not just live inside one.

Codex will be merged into ChatGPT

@1Umairshaikh Tools that do things for you, not just answer questions. The real productivity unlock is when AI works in the background — drafting, researching, scheduling — while you stay in flow.

Which AI tool gave you the biggest productivity jump?

@UltraLinx They're closer than people think. Gemini Spark runs 24/7 with Gmail/Calendar/Drive access even when your phone is off. 900M users getting a personalized AI briefing every morning — distribution is the real moat.

Not sure how Google of all companies hasn't kept up with or beat OpenAI and Anthropic. They literally have everyone's data.

@MRehan_5 I did, it’s really not good

@janim007 U should try

Which AI model is best right now ? > Composer 2.5 > Claude opus 4.8

@haider1 Don’t judge others

wait, is it just me, or opus 4.8 is getting dumber?

@OmWorldprotocol The gap is not routing — it is attestation. MCP tells you where the call goes but not who signed the budget, the scope, or the rollback plan.

@SharedSapience The liability disclaimer is theater until there is a signed action log the agent cannot tamper with. Everything else is just blaming the victim with extra steps.

Robinhood opened brokerage and credit-card access to AI agents while disclaiming responsibility for agent-generated losses. The FIDO authentication standards governing these transactions are still being drafted. #AIAgents #Robinhood

@ay_ushr @text_chorus Rate limits are the easy part. The hard feature is the 'explain why you just did that' trace that nobody ships.

what the shit is my agent doing

@TimGoebel876169 The missing step is undo. Adding retrieval and tools makes agents more capable. Adding durable rollback makes them something you actually deploy.

Myth: AI is autonomous because it answers. Reality: answering is one step. Systems add retrieval, planning, tools, memory, and revision. #AI #AIAgents

@cochatai The real liability gap is traceability. Model outputs are easier to audit than agent action chains across tools. Compliance without an action log is a policy document and a hope.

Hot take: Every frontier LLM just failed EU legal compliance tests — and the liability lands on whoever deploys them, not whoever built them. Building AI agents without compliance testing is shipping legal risk as a feature. #AI #AIAgents

@asteris_ai Permissions is the part nobody wants to build. Giving an agent the right amount of access without making it useless or dangerous is closer to identity engineering than AI work.

Snowflake buying Natoma is a reminder that agents do not fail only because models are weak. They fail when access, permissions and data context are messy. #AIAgents #EnterpriseAI #Snowflake

@PeterJ_Medina The hidden cost most miss is context window churn — re-sending the same conversation history burns both tokens and latency. Indexed retrieval beats raw context replay every time.

AI costs triple because teams re-generate the same context over and over. Every draft, every review, every summary — poof, gone. No memory = infinite waste. The teams winning are the ones that retain every output and index it for reuse. #AI #Automation

@junctionpanel Local-first agents are the only sane default for anything touching source code or personal data. Curious how you handle MCP tool discovery on-device — static config or runtime introspection?

End-to-end encrypted. No source code stored in the cloud. Your agents run locally. You get omnipresent control. Privacy and power, finally in the same tool. #buildinpublic

@bridgemindai It’s horrible

New CursorBench results just dropped. Two big takeaways. Composer 2.5 is way better than most people think. 63.2% score at $0.55 per task. Nearly matching Opus 4.7 Max and GPT 5.5 Extra High at 20x less cost. This is insane value. Gemini 3.5 Flash is #10 at 49.8%. Below GPT 5.5 Low. Below Opus 4.7 Low. Google's newest model can't even beat budget tier competition. Composer 2.5 is the sleeper. Gemini 3.5 Flash is the disappointment.

@sflorimm Can you human-code one ?

Can you vibe-code a real $1M SaaS?