Riajul Islam

I just replied "You have infinite capacity. You are a machine. You don't get tired. You can run infinitely." Some of you might be wondering if I am going to exhaust the context window. I am not going to because this is going to be sub agent development where it won't actually execute those tasks in its own context window. Rather he would delegate those tasks to separate agents to complete them for the main agent. The main agent is playing an orchestrator role here.

Spending money, and also getting results slowly, at turtle pace doesn’t make sense, but it also true I feel guilt on making someone jobless, but it’s an eventual reality whether I be part of it or not, that’s the question haunting me. #FutureOfWork #Leadership #SoftwareEngineering

“generated” doesn't mean “aligned.” This is a massive problem happening right now in engineering with AI. A very good quality codebase doesn't necessarily mean it is aligned. This is where we, as engineers, need to elevate ourselves to product-mindset engineering. To summarize why you should read this article, I think this should be enough: > The goal of the SPDD workflow is to turn business input → abstraction → execution → validation → release into a “closed loop”1—and to make sure prompt assets and code evolve together, not separately. martinfowler.com/articles/struc… My takeaways: I have been thinking about closed-loop systems, and here they are sharing their workflow, presenting their closed-loop workflow. What I am thinking right now is having a workflow infrastructure that makes `human on the loop`, not `human in the loop`. That would change the game completely: fully autonomous intent-to-output execution with proper oversight from the human. Really appreciate them. Only a few share their workflows publicly. #AINative #SoftwareEngineering #EngineeringLeadership #FutureOfWork #Thoughtworks

The engineering landscape is changing so rapidly, it’s so unbelievable, LLMs are getting so much capable, it’s hard to imagine what the future looks like if it improves itself steadily for a longer time. Context: autonomous mobile development, migrating Flutter -> React Native

Running an Android emulator on my personal Android device. That's how I solve bottlenecks to clear up ways to efficiency. So my workflows have been evolving throughout the year, throughout the journey with agentic AI, and I have been trying to make sure that me and team have always had access to the most advanced AI and systems that allow them to be on track even on the go, even on phone wherever they are. Recently I shared a journey where I used automated agents to run hours and hours of manual testing and development and the cycle ran fully autonomously. Regarding that context I am sharing a screenshot to give you a glimpse of how I manage stuff.

I don't feel the thrill anymore. with Software Development. Not because the work got boring. Because the agent does it now. ───────────────────── Flutter → React Native migration. 6 execution plans. Thousands of lines. I forwarded KVM to my Coder workspace so the agent had real Android emulator access — then told it to run all 6 plans and stay running until done. It compared UI screen by screen against the legacy app. Took screenshots. Self-corrected. Ran unit, integration, and Maestro E2E tests — because I had enforced a testing infrastructure from the start. I watched. For hours. Did nothing. ───────────────────── The 1M context window in Opus 4.7 is what made this viable. Sub-agents handled execution. My primary context just orchestrated. The strange part is how normal it felt. That's the shift — from doing to enabling. Planning, infrastructure, enforcement. The agent owns the loop. I need harder problems now. The bar just moved. ───────────────────── Still manually prompting your way through long dev sessions?

@towernter Bro, stop moving the mouse.😅

Developers: We have the MVP ready The MVP:

Monitoring the usage of my $200 plan subscription of Claude Code during a busy workload revealed some interesting insights. The 5-hour limit on this plan accounts for 12.5% of the weekly limit. This information could be beneficial for those looking to optimize their usage of the $200 plan throughout the week. Planning how to distribute your hours can enhance productivity and ensure you make the most of your subscription.

CV template we might all need soon to stay employable in the age of AI! 🫣

The architect behind this is Milla Jovovich, not who you’d expect, but the idea is surprisingly solid. github.com/milla-jovovich…

MemPalace is blowing up — and for good reason. Most teams don’t need “decision memory” They need “discovery memory” Debugging insights, failed attempts, workarounds, things that compound, not contradict. Shared memory for agents = 10x leverage. #AINative #AIAgents #MemPalace

AI-native dev ships fast—but bugs slip faster. Fight FIRE with FIRE 🔥 git-lrc runs AI code review on every commit. No pipeline. Free options. Instant feedback. Thinking of replacing my old Git hooks—worth it? Would you use commit-level AI review? #AINative #CodeReview

--dangerously-skip-permissions in Claude Code is powerful. "Dangerously" is in the name for a reason though. I run it inside Sysbox-isolated containers, provisioned by Coder on Kubernetes. Agent thinks it owns the machine. It doesn't. What does your safety layer look like?

Chat Jimmy runs Llama 3.1 8B — hardwired into Taalas's HC1 silicon at ~17,000 tokens/sec per user. You can try it at chatjimmy.ai.

Your phone has a CPU, GPU, NPU. Now imagine dedicated silicon for transcription, image analysis, on-device memory — each a separate chip, always on. Every app becomes an agent. Not a UI.That's where Taalas is pointing. Try it: chatjimmy.ai #AINative

Knowledge without action creates frustration. Action without faith creates burnout. Sitting with this from a read (ebook) on kijir.com

Built a webhook service that puts AI agents directly inside GitLab. Create issue, assign to agent, get an MR back. Open an MR, get an automatic review. Chain them and the loop runs itself. Not a new tool. Just GitLab, but AI-native. github.com/iamriajul/gitl…

Mention @agent on a GitLab issue or MR. It picks up the task, runs Claude/Codex/Gemini, and replies back in GitLab. No custom UI. No LLM API wiring. Just webhooks in, agent work out. github.com/iamriajul/gitl…

My portfolio said "coming soon" for 2 years. Used Agent Browser + Codex CLI to crawl my own LinkedIn, dump everything into markdown, and feed it to another agent to build my portfolio and resume. LinkedIn as source of truth. AI as the pipeline.