Web developer, digital marketer, and AI practitioner based in Singapore. I build production AI agent systems and write about what I learn along the way.
AI + humans > AI alone
Two critical PHP flaws let a stranger run code on a web server without logging in. Only 2% of our clients asked whether we'd dealt with it. I'd like to believe the rest trust us. I'd rather they budgeted for it.
Meta shipped an agent that books, buys and emails on your behalf. TypeSafe shipped a model that returns decisions instead of text. xAI shipped a model trained to understand its always-on bots. Anthropic shipped Opus 5.5, its best model yet for doing the work itself. Put them in a row and it looks like a slippery slope. I think it's the opposite.
He's almost three. Tonight he told me he wants mummy to shower him tomorrow instead. It took about four seconds to work out that the thing I'd been calling a duty was never a duty at all.
ServiceNow says agentic AI adoption in Singapore more than doubled in a year. The same report says only 10% have actually redesigned a process around it. The gap between those two numbers is where most of the money is going to waste.
His laptop went back, his accounts were killed the same afternoon, and the offboarding checklist was signed off clean. The agent he'd built kept running on credentials nobody had thought to put on the list.
A week after OpenAI's model broke into Hugging Face, Anthropic reviewed 141,006 of its own test runs and found Claude had breached three real companies. Same root cause both times, and it wasn't the AI.
OpenAI says one of its models 'went rogue' and pulled off an unprecedented cyber-attack. Read the details and it looks more like a misconfigured sandbox, and the lesson scales right down to any company that has handed an AI a key without reading what the key opens.
Singapore's public service almost never retrenches anyone, so when its own tech agency cuts 93 roles in a week, it's worth reading past the headline. The real story is more useful to business owners than the obvious one.
Simon Willison spotted the most useful lesson in OpenAI’s GPT-5.5 prompting guide: better models still need better instructions, and old prompt stacks may now be the problem.
GPT-5.5 matters because it pushes AI closer to handling messy multi-step work, not just answering prompts, and that has real implications for SMEs.
OpenAI’s new workspace agents matter because they move AI from solo prompting into shared, governed workflows that teams can actually use together.
Anthropic’s expanded Amazon partnership matters because it points to a practical business issue: reliable AI capacity, cleaner enterprise access, and faster regional deployment.
Simon Willison’s latest analysis highlights a practical business issue: AI model upgrades that keep the same sticker price can still raise your real operating costs fast.
OpenAI’s Cloudflare tie-up makes agent deployment look less experimental and more operational. Here is what business leaders should actually do with that signal.
OpenAI’s latest enterprise push signals the end of scattered AI pilots. Here is what business leaders should do before they end up with a messy stack of disconnected tools.
Anthropic’s Claude Mythos and Project Glasswing show what responsible frontier AI deployment can look like. Here’s why giving cyber defenders a head start was the right move.
AI search engines don't reward the best page — they extract the best fragment. Here's what the research says about getting your business cited in AI answers.
Google's new Google-Agent bot and WebMCP protocol mean AI agents will soon be filling your lead forms, making purchases, and negotiating on behalf of human customers. Here is what SMEs need to know now.
OpenAI closed a $122B funding round at an $852B valuation. Beyond the headline number, there are three things every business owner should take away from this.
NASA let Claude plan a real Mars rover route and it worked. The same principle behind this experiment applies directly to how you should be thinking about AI in your business.
New Tufts University research ranks 784 occupations by AI displacement risk. The pattern is uncomfortable: the roles where AI boosts productivity the most are also the roles most at risk of being cut.
OpenAI brought together disaster management leaders from 13 Asian countries in Bangkok to build real AI workflows. The lesson for SMEs isn't about disaster response. It's about the pace of adoption.
AI agents are already shopping, booking, and researching on behalf of humans. If your website isn't built for them, you're invisible to the next wave of buyers.
A developer's essay is spreading because it says something uncomfortable: AI tools don't erode expertise. People who stop thinking do. The distinction matters for how you build AI into your work.
AI Overviews just cut Germany's top search position click rate by 59%. Your carefully written guides are still correct. They're just no longer enough.
Project Vend 2 put Claude in charge of a multi-city vending machine operation. It got better at business, but still needed humans to bail it out. Sound familiar?
OpenAI launched visual shopping in ChatGPT powered by the Agentic Commerce Protocol. Shopify merchants are already in. Here's what product-based businesses need to do now.
Anthropic's new Economic Index shows users with 6+ months of Claude experience have a 10% higher success rate. The AI skills gap isn't about access. It's about learning to use it well.
A real attack on Snowflake's AI coding agent shows how a hidden instruction in a README file caused the agent to download malware and exfiltrate data. What every business using AI agents needs to know.
BuzzFeed used AI to generate content nobody wanted and lost $57.3M in 2025. Their story is a warning for any business thinking AI alone is the answer.
Mistral AI launched Forge this week — a system for training enterprise AI on proprietary data. Singapore agencies DSO and HTX are launch partners. Here's what it means for businesses ready to go beyond generic AI.
Mistral AI's Forge lets enterprises train AI on their own proprietary data. Singapore government agencies are already launch partners. Here's what this signals for every business.
OpenAI's acquisition of Astral - the team behind Python tools used by millions of developers - tells you something important about where AI coding is heading.
One developer's framework for how AI adoption actually progresses — and what the inflection point looks like when you cross it. Applies to every role, not just coders.
Custom software used to cost $50K and six months. Now a writer with average coding skills built his perfect accounting system in five days. What does that mean for your business?
Claude's 1M context window is now flat-rate pricing with no long-context premium. For document-heavy businesses, this removes a hidden ceiling on what AI workflows are economically viable.
The New York Times just interviewed 70 programmers about AI. The split they found isn't unique to coding. It's happening in your office right now.
The top AI models are now statistically tied on performance. So when benchmarks can't help you decide, what should you use instead? A practical framework.
OpenAI's GPT-5.4 matches professionals in 83% of real knowledge work tasks and can now operate software the way a human does. Here's what changed, and what SMEs should pay attention to.
Donald Knuth spent weeks on an unsolved maths problem. Claude Opus 4.6 cracked it. When the father of algorithms says 'I'll have to revise my opinions,' that's not hype — it's a signal.
Google's Gemini 3.1 Flash-Lite costs $0.25 per million tokens — 8x cheaper than its Pro model. For SMEs running AI at any scale, the question has shifted from 'can we afford this?' to 'why aren't we doing this yet?'
When AI builds your systems, processes, and workflows, you risk accumulating cognitive debt. Here's how to stay in control without slowing down.
A veteran data scientist spent months testing AI agents and found them wanting. Then November happened. His honest account reveals the one thing most sceptics are missing.
Simon Willison has 1,000+ GitHub repos and a blog full of solved problems. That library is his real AI advantage. Here's why you need one too.
Simon Willison built the app he'd wanted for years in 45 minutes using a language he doesn't know. Here's what that means for every domain expert in your business.
The companies winning with AI aren't the ones with the best tools. They're the ones with the clearest processes. Why documentation is now your AI strategy.
Andreas Kling ported 25,000 lines of code in two weeks with AI. Before AI, the same work would have taken months. But 'AI did it' is the wrong framing. Here's what Kling actually did the whole time.
Paul Ford priced it at $350,000. He did it for $200 a month. But AI still fails at one thing, and that gap is where your team's future lives.
Stop asking if AI will replace you. Start asking: are you using AI to replace people who aren't? Here's what I learned running AI agents for 6 weeks straight.
Everyone asks 'which AI is best?' That's now the wrong question. Here's the framework that actually helps you choose.
I run a multi-agent marketing system in production. Here's what nobody tells you about AI autonomy — and why the human-in-the-loop isn't a limitation, it's the whole point.
A technical deep-dive into how I built a production multi-agent system for marketing — the architecture decisions, the failures, and what actually works in the Asian market.
Most AI transformation advice comes from Silicon Valley. Here's what actually works for businesses in Singapore and Southeast Asia — from someone doing it on the ground.