Every few months someone asks me which AI tool they should switch to, as if the answer is a football transfer. My honest history looks like this: ChatGPT, then the paid tier, then Gemini, then NotebookLM, and these days Claude, Claude Desktop and Claude Code. But the word “switch” is wrong. I drift to a new platform when I have exhausted what I can usefully get from the current one. That does not mean the old one gets thrown out.
Nothing got deprecated
Today the tools run as one working ecosystem, each doing what it is best at. Gemini handles my image generation, alongside Project Reverie, the Byteplus-based studio we run at OpenMinds. ChatGPT and its newer image model is my direct fallback when Gemini hallucinates. Claude is my full-time assistant, holding individual projects in Claude Projects, and Claude Code is where working proof-of-concepts and internal tools get built. Nothing was replaced. Everything settled into a role.
What Claude Code does for me
Two things, broadly. The first is websites: creating and updating WordPress sites through MCP connections, work that used to mean hours of clicking through admin screens. The second used to require a full-stack developer: functional applications. The clearest example is our internal social tracking dashboard. Managing multiple clients and accounts across multiple channels is a monitoring headache, so the dashboard pulls it into one place, produces an AI summary every week, and generates a full PPTX report with one click. It also runs complete SEO and GEO audits, recommends the changes, and with the right skills wired in, deploys them.
Where the human stays in the loop
The workflow is not fire-and-forget. You cherry-pick the skill for the job, then review the result against the actual, real-life implementation to make sure nothing has drifted or been misunderstood. Human judgement stays mandatory wherever nuance lives: the client’s brand, and the decisions that were made in rooms the AI has never been in.
My personal guard rail is simpler and stricter. I still go through what AI produces for me line by line, and even in Claude Code I want to understand why it suggested something before I implement it. I would not say there are tasks I refuse to hand over. It is more that AI earns its place by complementing my work, uplifting the foundational findings, safeguarding the research, and staying grounded in real-life understanding rather than hallucination. The moment I stop checking is the moment that stops being true. I go deeper on where that line falls in client work in how I actually use AI in my consultancy.
The wrong way to use it
The most common mistake I see is treating AI like a search engine. You type a question, take the answer, and move on. That is validation, not uplift, and for me it is the wrong process entirely. Used properly, AI is like having someone reliable across many types of task and expertise at once, and the efficiency it returns means you can simply do far more today than in the yesteryears.
In Malaysia there is a second problem: people using AI at surface level while marketing themselves as gurus, with content that is mostly referenced from elsewhere. The market is saturated with self-described experts. I will not add to the pile. I do not consider myself an AI expert. I use AI to my benefit as much as I can, to assist, to uplift, and to maximise my efficiency, and I only ever recommend what I have already run in my own work.
