Ideas for running a sharper digital operation.
Practical write-ups on web development, SEO, AI search and how we work with Singapore SMEs.
Pydantic AI for Building AI Agents: What It's For and When It's Worth Using
Pydantic AI brings the type-safety and validation discipline of Pydantic to agent building. Here's the problem it actually solves, how it compares to the alternatives, and whether it holds up in production.
AI Frameworks vs AI Toolkits: What Actually Sets Them Apart
LangChain, CrewAI, LlamaIndex, and the Vercel AI SDK all get lumped together as "the AI stack." They solve different problems. Here's what an AI framework is actually for, and where a toolkit like the AI SDK fits instead.
Where Langfuse and Harbor Fit Into How We Evaluate Our Own AI Agents
Our own self-monitoring pass catches an agent quietly going off the rails. It was never built to grade whether an answer was actually good, or to prove a change is safe before it ships. That's the gap Langfuse and Harbor fill.
Under the Hood: How an AI Agent Actually Turns a Sentence Into a Real Action
An LLM has no hands and no terminal access. Here's the six-step pipeline, registration, generation, interception, parsing, execution, and feedback, that turns a prediction engine into something that can safely take real action, and why we build the execution step narrower than most.
Adding a New Capability to Our AI Agent Is a Two-File Change. Here's Why That's the Point.
When a client asks for one new thing their AI agent should be able to do, the honest answer to "how long will that take" depends entirely on how narrow the underlying tool interface is.
What Actually Happens When You Send a Message to One of Our AI Agents
A trace of one message from browser to model and back, including a subtle streaming-protocol bug we found and fixed, and why almost none of this is visible from the chat window.
One Backend, Many Products: How a 'Skill' Lets Us Ship a New AI Agent Without a New Backend
A tutor for secondary school students and a marketing brainstorming tool run on the exact same reasoning loop. Here's the abstraction that makes that possible.
What It Actually Takes to Build an AI Tutor You Can Trust
AI in education gets discussed as either a replacement for teachers or a cheating machine. The actual question worth asking is narrower and more useful: what makes one trustworthy?
How AI Agents Are Changing Marketing for Singapore SMEs, Without Replacing Your Team
The useful version of AI marketing isn't autonomous posting, it's grounded ideation and disciplined testing that makes a small team's existing content instincts go further.
Harness vs Agent: The Questions Your AI Vendor Should Be Able to Answer
Every AI vendor's pitch sounds the same once you strip away the buzzwords. Here are five questions that actually tell a serious build apart from a demo that hasn't been stress-tested yet.
How We Watch Our Own AI Agent for Signs It's Gone Off the Rails
"It ran without crashing" and "it's actually behaving well" are different questions. Here's the self-monitoring pass we run over our own AI agent to answer the second one.
We Gave Our AI Agent Its Own Database. Here's Why That's Safe.
Letting an AI agent touch a database sounds risky. Scoped correctly, with a hard read/write split and zero access to real business data, it's actually what makes multi-step tasks reliable.
Giving Our AI Agent a Memory That Doesn't Grow Forever
Long-term memory sounds simple until it actually grows. Here's how our AI agent remembers across sessions without either blowing its context budget or quietly forgetting something it needed.
Why Our AI Agent Asks Before It Acts, and Never Acts Twice
Two failure modes matter more than raw model capability: an agent that acts without asking, and one that accidentally acts twice. Here's how we solved both at the infrastructure level.
AI Agent = LLM + Harness: The Equation Behind Every AI Product We Build
Everyone says "AI agent" and means something different. Here's the equation we use to cut through it, and why we built our own harness instead of wrapping a chat window around a model.
Why We Don't Default to OpenAI for Every AI Feature We Build
Most AI development tooling assumes you'll use a handful of major providers and stops there. Here's what it actually takes to plug a different model, like a Southeast Asia-trained LLM, into a real product instead.
What Happens When You Give an AI Agent a Memory and a Schedule
A stateless chatbot forgets everything the moment the conversation ends. We built a persistent AI agent that remembers, runs on its own schedule, and checks in without being asked, then looked at where that pattern actually earns its keep in a business.
Teaching an AI Agent to Actually Manage a Calendar
Scheduling is simple to describe and tedious to actually do. Here's how we built an AI agent that reads a plain-language request and creates the calendar event itself, instead of just describing what it would do.
How We Built a Multilingual AI Assistant Using n8n and a Southeast Asia-Trained LLM
Most off-the-shelf AI assistants default to models trained mostly on English, Western data. Here's how we wired a Southeast Asia-trained LLM into a Telegram assistant using no-code automation instead of a custom backend.
Why We Test Every AI 'Skill' Before It Touches Client Work
Most teams bolt AI "skills" onto their coding agents and trust the output. Here's why we test ours first, and how we use AI agents day to day.
Why Most AI Agents Break the Moment You Actually Rely on Them
A slick AI demo and an AI system your business can actually depend on are two very different things. Here's the gap most vendors don't mention.