Malhar Ujawane

I build agentic products in public. The interesting part of AI is not just what a model can say. It is what a system can do reliably. My background: - PhD in Computer Science - ex-Walmart Staff Engineer Here I write about: - agents in production - system design around LLMs - local vs frontier model tradeoffs - failure modes, retries, memory, trust If you care about practical AI systems more than demo theater, welcome.

@irushi Codex/Kimi K2.5 seems like the best choice for OpenClaw, @AnthropicAI really screwed up this time

Yeah I'm cooked, going to burn through $200 in the first 12 hours itself... not going to make it through Time to find an alternative, ty @AnthropicAI you've been awesome <3 but now Ares needs another fuel for his spark (I still have api access and rely on claude code and opus for everything else so by no means will I be ending my subscription - just cutting down to just 1 main account and route the other max accounts to API expenses and use it a lot less than I was before and add local models + openAI + anthropic to get the best output - but nothing will replace the golden era of max <3 RIP)

@andrewfarah Love this project!

sharing my first open source project a CLI for downloading and syncing your X bookmarks locally so your agent can access them. it's free › npm install -g fieldtheory › login to your X account in a chrome tab › ft sync (done!) bonus: › ft viz › ft classify

People are focusing on the wrong part of the Claude Code leak. The most important unannounced feature may be “dream” — a periodic memory consolidation pass. And yes, that is much more interesting than basic “memory.” Why? Because memory by itself is cheap. Any agent can keep notes, append logs, or dump facts into files. That does not mean it understands anything over time. Raw memory usually turns into: - duplicate facts - stale assumptions - contradictory notes - bloated context - worse retrieval over time In other words: memory rot. “Dreaming” appears to be the layer that fixes that. Instead of only storing new information, the agent periodically reviews what happened, extracts what actually matters, updates durable memories, removes bad or outdated facts, and keeps the memory index clean. That is far more powerful than ordinary memory. Memory says: “I saved everything.” Dreaming says: “I figured out what was worth keeping.” That is the difference between a notes app and a system that can build continuity. Bigger context windows help. Better retrieval helps. But agents that can rewrite their own memory intelligently will be in a different class. If this is real, Claude Code may be pointing at the next serious leap in agent design: not just persistent memory, but self-maintaining memory.

Finally a way to reuse my email account from the Orkut days

@charlieholtz Congratulations man!

Big news for @conductor_build! We've raised a $22m Series A from Spark and Matrix. We raised this round from @ilyasu at Matrix, who also led our seed round and is joining our board, @nabeel at Spark, @ycombinator, and founders of Notion and Linear. We're grateful to be working with investors we trust and admire. Here’s how we got here and where we’re going:

@DJIGlobal Taj Mahal

Giveaway time! Here's your chance to win the DJI Avata360 (DJI RC2) — DJI's first panoramic drone, redefining aerial creativity with flagship-level 8K HDR imaging. How to enter: 1. Follow @DJIGlobal 2. Like and share this post 3. Bonus Chance: Comment below — what's the one place you'd love to fly through and capture in full 360°? · Time period: 2026/3/27 - 2026/4/27. This giveaway is not affiliated with or endorsed by X. One winner will be selected at random through a third‑party platform. Good luck! Video created by: aleixalbet

@trq212 During peak hours Claude Code just stops working with 4.6 Opus on Max plan. Any resolution for this?

To manage growing demand for Claude we're adjusting our 5 hour session limits for free/Pro/Max subs during peak hours. Your weekly limits remain unchanged. During weekdays between 5am–11am PT / 1pm–7pm GMT, you'll move through your 5-hour session limits faster than before.

@thsottiaux Also can we fix the Quit feature where it keeps asking for confirmation to close even when no threads are running?

Codex deserves great tools. We are about to expand its toolkit a whole bunch and I can’t think of using anything else anymore for all my daily tasks, way way outside of coding.

You can now use Claude Code directly within Obsidian with Claudian plugin. No more juggling between Obsidian and Cursor

@charlieholtz Been using the new sidebar UI since last few days, its so much better than the previous sidebar. Could you add a archived chats section to review past work without needing to keep the worktree in memory just the chats?

Introducing our new, simplified sidebar! You can now see GitHub status at a glance (merged, failing CI, conflicts) Hover any workspace to see metadata and get easy access to next actions without leaving your chat:

Things OpenAI has sunset so far: - Dall E - Operator - Sora - Chat Completions Endpoint What’s next?

Working on building a AI Agent Orchestrator for Developers to replace Claude CLI and Codex. Would you like to try it?

@usenaive @OfficialAmogh Naive

Introducing Naive - hire autonomous employees with their own identity. Own compute. Own bank account. Own legal entity. Own email. Own credentials. Own mobile. No humans-in-the-loop. They sign up for tools, pay for services, deploy apps, file documents, and run your entire company. Describe a business. Naive runs it. Reply "Naive" + RT. Get $100 credit for free.

@tannerlinsley @conductor_build is the answer

Ghostty was fun, but time for something else. I still love opencode, too but with CC plans dead on it… I’m feeling lost. Full GUI? T3 Code? Opencode GUI? Warp? Back to cursor? Try CC again? Raw Codex? My 🧠 hurts and I just need to keep shipping.

@reach_vb Can't wait to attend these!

Codex 🤝 India! So so excited about bringing the magic of Codex to India 🇮🇳 See some of you around India, soon!!

@NebulaAI @atulkumarzz Down!

Introducing Virtual Computer We're giving you a persistent computer with built-in agents that use the web, develop software, and get smarter over time. Create and deploy a website. Generate and edit videos. Automatically fix bugs as they get reported. Your AI team works 24/7 to achieve your goal. Reply with what you'd build - gifting $50 in credits to the most interesting answers

@hamptonism why does it look like boosters from a Mini Starship @elonmusk?

Assembling a Quantum Computer:

@ElaraJordan github.com/justmalhar Will charge less if it’s just 2 hours

I have a $2000 budget for someone to vibe code me an openclaw skill. For the right person, this should be like ~2 hours. I tried upwork but there is some clever prompt engineering. Ideally you're a native english speaker. MUST SHOW YOUR GITHUB and have at least something public

The real bottleneck in many AI products is not generation quality. It’s trust architecture: what can run automatically, what needs review, and what should never be silent.