The 12-Month Clock
A data scientist's reflection on AI's exponential rise. Originally published on LinkedIn, 27 November 2025 — republished here in full.
Some thoughts on the back of a lot of reflection over the past 24 hours.
The Timeline We Keep Forgetting
November 2022 — ChatGPT bursts onto the scene. It can't do math. It can barely code. No reasoning capabilities. Constant hallucinations. Most of us dismissed it as a parlor trick.
2023–2024 — Multi-modalities get introduced. LLM chains become a thing (this was when I first started using LangChain and LlamaIndex). RAG creates a surge in intelligence and grounding. Things start getting interesting.
Early 2025 (February) — Co-pilot agent releases on VSCode Insiders. This was the first AI agent I ever used. And honestly? It could barely put together a working frontend across multiple prompts. It really struggled to write a backend on any scale bigger than a few functions and API endpoints. It couldn't deal with data or databases to save its life.
(Side note: Claude Code was actually in research preview at this point, and tool-use was becoming adopted despite being theorized back in late 2023. MCP had been released just three months earlier.)
May 2025 — Sonnet 4 releases. Claude Code integrations mature. MCP servers are on the rise, bridging the gaps between data and AI agents.
November 2025 — Where we are now.
Where We Are Now
Non-technical people can build and deploy small apps to help with their daily work.
The best agents and models can work relatively autonomously on a task until completion.
Interacting directly with Google search is becoming obsolete.
We are being forced to abstract our thinking to higher levels every 2-3 months.
Think about that cadence:
- AI can't write a file? Oh, now it can.
- AI can't write an accurate diagram? Now it can.
- AI can't manage a codebase? Now it can.
- AI can't deploy an app? ...Now it can.
The Remaining Walls (For Now)
The key limitations that still exist surround broader context, system design, architecture, scaling, future implementations, and governance.
AI can't design and manage an enterprise, production-level codebase?
Not yet.
But how long?
The Math That Made Me Think
Look at November 2022 to February 2025. That's 27 months of progress.
Now look at February 2025 to November 2025. That's 9 months.
The growth is exponential.
In those 9 months, we went from not being able to write a comprehensive file in a codebase to almost one-shotting an entire (albeit small) application.
This weekend, it took me less than 2 hours from conception to deployment for a relatively small CRUD app—fully deployed on Supabase and Vercel. No bugs. Scalable code.
Two hours.
It's worth asking: will AI be able to design full systems and manage projects autonomously in the next 12 months?
A Question Worth Asking
Are we being reactive instead of proactive?
I've been doing a lot of thinking about my career trajectory lately. Not from a place of fear, but from a place of genuine curiosity about where the opportunities are heading.
I feel like the next 12-18 months represent a real inflection point. And I want to position myself well for it.
A (Maybe Controversial) Take
Here's where I land after all this reflection:
We shouldn't be focusing on what AI can't do.
But here's the part you might not expect:
We also shouldn't be focusing too much on what AI can do right now.
I think we should be focusing on what AI is most likely going to be able to do in a 3-4 month lead time—and prepare for that.
The Questions I'm Now Asking Myself
Why would I build a locally-running, standalone sales agent? Why wouldn't I build this on an enterprise platform like Azure?
Why am I trying to solve niche little problems instead of putting my energy into learning what the hyperscalers are doing and the path they're going down?
Because they hold the key for security and governance. And that's where the real scale lives.
In 3-4 months, is it possible that the Data Scientist role evolves into an agent orchestration role on hyper-scaling platforms?
I genuinely don't know.
But I do know this: every time I think there is something that is years away, it turns out to be only months away.
And every hour I spend learning and preparing for this feels like time well invested.
The Takeaway
We're not preparing for a future that's coming in 5 years. We're not even preparing for next year.
We're preparing for what's coming in 90 days.
And then the 90 days after that.
And the 90 days after that.
The question isn't whether AI will transform our roles. It's whether we'll be the ones shaping that transformation—or playing catch-up.
The opportunity is now.