Hi, I'm Aakash.
I build backend systems and production AI applications.
Recently built:
• Multi-agent AI platform
• LangGraph workflows
• RAG pipelines
• Java microservices
Currently looking to build challenging AI products.
GitHub : https://t.co/4BDfADduMY
@MauryaAlka76391@MauryaAlka76391 ye old video hai and esi ke karan ye jail me hai , aapka info system lag hai aapko lag rha ye abhi bola hai 😀 Aapki research team ko thora work Krna chahiye, didi aapka arguments invalid and biased hai , u speak without evidence, first do proper research
We ran an agentic system in production for 6 months.
5 things nobody warned us about:
1 80% of the work is observability, not the model
2 Context window ≠ memory
3 Your best prompt won't survive a model update
4 Users don't want autonomy, they want undo
5 The bottleneck is trust, not capability
The agentic demo: "Book me a flight."
The agentic reality: 40 tool calls, 3 retries, one hallucinated airport code, and a $2,000 charge to the wrong card.
The gap isn't intelligence. It's error recovery.
Ship agents that fail gracefully or don't ship them at all.
Everyone's building "autonomous agents."
Almost none of them are autonomous. They're a prompt chain with a for-loop and a marketing budget.
The teams winning right now aren't the ones with the most agents. They're the ones with the best evals.
Reliability > Autonomy
Coding agents are making first iteration of software engineering faster. They’re also making weak engineering easier to ship. Production doesn’t care how quickly the code was generated. It cares about reliability, failure modes, security, and who debugs at 2 AM.
The bottleneck is shifting:
1 Writing code is getting cheaper.
2 Judgment and customized quality code are getting more valuable.
Better engineers won’t be those who type every line. They’ll be the ones who design systems, build evals, and own reliability. AI-native development shouldn’t mean blindly accepting generated output. It should mean tighter loops: generate, test, observe, improve.
In the age of agents, code generation may be abundant. Engineering responsibility will not be.
What do you think: will AI raise the bar for engineers—or lower it until production failures raise it for us?
#AIEngineering #SoftwareEngineering #AgenticAI #LLMOps #SystemDesign #GenerativeAI
@manucastilloll Love this energy. I’m Aakash—building at the intersection of AI and backend systems from India. Always happy to connect with fellow founders and swap notes on what we’re building.
@AstasiaMyers Exactly. In the creator-software era, distribution isn’t just a growth channel—it’s part of the product itself. The winners will combine strong technology with trust, taste, and a built-in audience.
@insanekrishnaa DSA helps, but I wouldn’t make it the only focus. If your goal is strong roles, pair it with real projects, backend/system design, and a clear target company/role
@bhowconda Great breakdown — the key insight is that the control plane is really about normalizing protocols and preserving event semantics, not just wrapping multiple agents in one UI. The distinction between PTY, ACP, and native integrations makes the architecture tradeoffs very clear.
@lkronhubbard Such a great reminder that the best product work is often the invisible kind. Huge credit to the team for making the hard parts feel effortless.
@AroraBhavyam@nvidia This is a great non-dilutive path for early AI startups. For someone building agentic systems on a lean budget, the DGX Cloud + H100 access alone is huge. Do you know if solo-founder setups (1 dev + MVP site) from India typically get approved, or do they prefer US entities?
@PrinceProjectX Congrats on the launch, Prince! Love the focus on community-building. What’s the tech stack behind Twittoo, and are you planning any AI features (e.g., smart feeds, content moderation, or creator insights)? Would love to follow the journey.