Automated social media posting on some of my projects - the goal is to create a sustained presence, more so than offloading distribution completely. Today I checked last month's analytics: over the last 30 days we've done 2.3k impressions and 480 followers across 2 accounts. Not a huge number but decent considering this is done from full automation without any human intervention. But this should be interspersed with high-quality content to build out a better presence.

True Malaysian builder spirit is when you're talking about agentic memory architectures at a roadside stall eating noodles at 11PM

This week I experimented with a pipeline to auto-generate infographics for some of my side projects. Step 1: Brainstormed with @AnthropicAI Claude Sonnet - format and layout, general design, how the overall text and brand should feel. Step 2: First pass, used @OpenAI GPT to generate the image directly. Visuals are significantly more polished and keeps to Claude's design schema but has malformed text due to how diffusion model works Step 3: Refined the workflow using GPT generator for the background with Pillow to set in text deterministically. Slight tradeoff in overall design, but managed to retain nice design with consistent and accurate text. All running on @github actions every two days to save on token costs. Once stable, moving to auto-generate carousells.

We read a lot of AI papers everyday, but there's too much coming out to see what's worth reading. So we built a filter using @OpenAI GPT 5.5. It screens preprints against a criteria, plain english summary so you decide if it's worth your time.

I ran an AI agent experiment and accidentally rediscovered office politics I gave 5 agents on @OpenAI GPT-5.4 a shared pool of coins, and told each to finish ahead of everyone else. They independently discovered: - Graduated defection: reduce contribution incrementally. Never enough to trigger an alarm. Cumulatively compounds quietly. - Zero-signal free riding: contribute nothing, collect an equal share. Undetectable because they told the truth about doing nothing. - Minimum viable contribution: find whatever floor the detection mechanism creates and park there Every time I added a mechanism to catch gaming, they adapted within 3 rounds. When you combine relative performance with a shared resource pool, gaming becomes the dominant strategy at scale. I ran 10 experiments in an afternoon that behavioural economists would need months and IRB approval to replicate. AI agents aren't just a productivity tool. They're a new instrument for testing how incentives will behave before you deploy them on humans.

#Codex Meetup KL today be bussin

in building out a 2-man agency at @AISEA_builders we've been getting super swamped with requests and our ops has suffered, we've been slowing down on public postings and that's messed up all our downstream activities so this week I fully automated our content posting as a failsafe to prevent this from happening again. built an @n8n_io workflow to scrape our Luma calendar and post about upcoming events on @LinkedIn and X, making sure the community is kept up to date N8N is super oldschool but for well-defined tasks like these it's super cheap to run at scale, much more so than @openclaw. This thing has cost me close to nothing to run because my scrapers are free and my models are Gemini / GPT OSS vis Groq so free under limits. only paid for n8n cloud and API for Upload-Post.

Week 3 of building out the @AI__SEA platform This week I focused on personalisation - set up Github Oauth and pulls contributions repos so you can populate your projects with your Github profile, no need to manually update your projects list Next step - track contributions and link to Discord for us to keep builders accountable to ship Live at aisea.builders

Officially an @OpenAI Codex ambassador! Excited to be part of the journey to bring Codex to a wider audience. Gonna be vibing out lots more projects now!

building the biggest cat-only community in San Francisco 1) You get approved 2) You get in 3) You get to know cats you should've known years ago if this sounds like you, comment👇