The Shift #1: Original Thinking in the AI Era
In a world where AI has democratised competence, 'good enough' is no longer sufficient. To survive and thrive, we must cultivate original thinking and a strategic business mindset—becoming the outlier in the dataset.
Important
Introducing: "The Shift" series
We are currently navigating a structural break in how technology is built and who is valued for building it. The Shift is a series of my personal observations and notes from the front lines of AI Research and Engineering. It explores how we must adapt, and examines the new archetypes required to survive a world of infinite, instant competence.
We are living through a strange paradox.
On one hand, AI has democratised competence. With a decent prompt, anyone can write a competent email, code a basic app, or even initiate a startup. The "floor" of human output has been raised significantly.
But there is a trap hidden in this convenience. Because AI models are trained on the sum total of human knowledge, they function by predicting the most probable next step. They are, by design, engines of consensus. If we all use the same tools to ask the same questions, we will all receive the same average answers.
In a world where "good enough" is free and instant, I'm realizing that original thinking and a ruthless business mindset are no longer just nice-to-haves—they are the only defences against becoming obsolete.
The Death of the "How", The Rise of the "What"
For the last two decades, the tech world rewarded technical execution—the "how." How do I centre a div? How do I optimise this SQL query? How do I deploy this to K8s?
Even just two or three years ago, my work as an ML engineer was dominated by questions: How do I fine-tune this SOTA model? How do I fit a 70B parameter model on a L4 GPU? How do I customise this CNN architecture for limited data?
But AI is rapidly solving the "how." This shifts the value accrual to the "what" and the "why."
Note
Even AI/ML Engineering jobs are not safe.
On Dec 4 2025, HuggingFace gave Claude ability to do full lifecycle of model fine-tuning. https://huggingface.co/blog/hf-skills-training
It takes a plain English command, picks the right hardware, writes the script, submits the job, and trains the model while you sleep. It supports modern fine-tuning methods (SFT, DPO, GRPO), converts models to GGUF for Ollama, and handles monitoring automatically.
---
Another example: "Proactive Agents" from Google Labs: https://www.youtube.com/watch?v=v3u8xc0zLec
Proactive AI does not even wait for your prompts. It analyses your codebase, and suggests the next steps.
This is where Original Thinking enters the equation. When the barrier to building anything drops to zero, the value shifts entirely to deciding what to build.
If you ask an LLM for a solution, it will give you the aggregate of every pattern that has already worked. It frames problems based on historical data. It cannot hallucinate a net-new paradigm shift—like the original Transformer architecture that replaced RNNs—because the consensus of the time was to build deeper, more complex recurrent networks, not to discard recurrence entirely. That leap required an intuition, contrarianism, and a distinct human perspective that looked "wrong" to the established trend.
To win in the AI era, we have to be the outlier in the dataset. We have to provide the spark that the model can't predict.
The Filter for the Flood: Why Business Logic Wins
If original thinking provides the spark, the Business Mindset builds the engine.
It is natural to feel a rush when you first start using AI. We have all been there—staring at a screen, amazed that a prompt just wrote a working app in seconds. That excitement is the fuel, but it isn't the destination.
The trap is mistaking that initial magic for a business model. It is now trivial to generate endless code or content, but in a world of infinite supply, volume is not value.
Fun Fact
The "Volume Trap" affects science, too.
This system of rewarding sheer output is creating a crisis in academia, where the flood of "vibe coded" papers is burying genuine insight: The Guardian | Artificial intelligence research has a slope problem, academics say: It's a mess
This is why the Business Mindset is now an essential skill for engineers. It acts as a filter. When you can build anything in minutes, the most valuable skill is knowing what is worth building.
Having a business mindset in the AI era means understanding that technology is just a lever, not the product.
- Does the math work? Just because you can automate a service doesn't mean people will pay for it.
- Can you sell it? When content is infinite, the person who understands psychology, trust, and distribution wins.
- Are you building or just coding?
- The Old Way: An engineer's value came from implementation details and 'making it work'.
- The New Way: An engineer's value comes from outcomes. Using AI to code faster is just efficiency. Using AI to identify a user pain point and ship a full-stack solution in a week is leverage.
The New Archetype: The "Cyborg" Strategist
The winners of the next decade won't be the people who hate AI, and they won't be the people who let AI do their thinking for them. They will be the Cyborgs.
These are the people who treat AI like the world's best intern: it's a tireless researcher and a rapid prototyper. But they never let the intern run the company.
- The AI brings the raw materials, the summaries, and the drafts.
- The Human brings the taste, the strategy, the ethics, and the "crazy" ideas.
The Bottom Line
AI has commoditised the middle. The center of the bell curve is now very crowded and very loud.
For me, the goal is to get comfortable at the edges. To cultivate a mind that thinks in ways an algorithm finds statistically unlikely, and to build the judgment to turn those weird, original thoughts into something real.
It's a reminder to myself: Don't just prompt. Think.
Tip
Key Takeaways for the AI Era:
- Consensus is a trap: If an AI agrees with your strategy immediately, it might be too generic.
- Curate your inputs: Original output requires original input. Read old books, have strange conversations, and experience the physical world to feed your brain data that isn't on the internet.
- Focus on leverage: Use AI to multiply your judgement, not to replace it.