The model is the supercar; your knowledge base and workflow are the road beneath it

What this article is about

When a new generation of strong models like Claude Fable and GPT-5.6 Sol arrives, the first question most people ask is: should I upgrade now? Have I already missed the boat? I look at something else first. How far have you organized your knowledge base, workflow and knowledge flow? Put the same supercar on different road surfaces and it drives like two completely different cars.

Fable and GPT-5.6 Sol both point to the same thing

Anthropic's Claude Fable and OpenAI's GPT-5.6 Sol take different approaches to capability and delivery. What they have in common is that the models are starting to handle longer, more complex knowledge work, research and tool use.

As models keep getting stronger, the documents, rules and workflows you have prepared will shape the final experience even more directly. Give two people the same strong model, and one can plug it into long-running tasks while the other can still only use it for one-off questions and answers.

Official information:Claude Fable 5 ↗　OpenAI GPT-5.6 ↗

One supercar, on three kinds of road

When the automobile was first invented, its speed and reliability did not immediately beat the horse-drawn carriage. Some people only saw the comparison in front of them; others started to see another possibility: cars would keep improving, and the roads would change along with them.

When the road is still mud and gravel, there is little felt difference between a slower car and a faster one. Once the surface becomes asphalt, a car's stability and speed start to show. And only when the highway appears does a supercar really have room to open up.

Dirt road: data is scattered, the process lives in your head

You can still use AI on a dirt road. General Q&A, one-off edits and summaries can still be done quickly. The trouble starts when tasks get longer, data piles up, and the judgment calls get more complex.

• The same project has several versions, and you don't know which one is the latest.
• Your way of working exists only in your head, never written down as steps.
• Every time it starts, the AI has to ask again about the background, audience, format and constraints.
• Data is scattered across LINE, email, cloud drives, chat logs and assorted note-taking tools.
• There are no acceptance criteria when you finish, and new lessons never get written back into the original documents.

Here, the model can only guess its way across loose ground. A stronger model does guess right more often. But without the data, rules and goals, it still doesn't have enough to work with.

Asphalt road: documents centralized, the workflow starts to settle

The key to the asphalt road is to spare the AI from having to understand you from scratch every time.

At this level, articles, presentations, meeting notes and research reports start to form stable workflows. When you upgrade the model, the difference is easier to feel, because the task already has a clear route.

Highway: knowledge, rules, permissions and acceptance all connected

Once these structures are connected, the AI has a chance to move from answering questions to taking over a whole stretch of work.

My documents are my system

I started using Obsidian in 2024 and, over time, worked my way up to Harnessing Engineering and Loop Engineering. That journey has left me more and more sure of one thing: tools can be swapped out, but your own documents have to keep accumulating.

If your documents use common formats like Markdown, plain text or CSV, the cost of moving is usually lower. Switch to a different AI, a different Agent or a different note-taking tool, and the content you already organized can still be used.

Six steps to pave a dirt road into one AI can run on

This is what I call Loop Engineering: input, execution, acceptance, correction, recording, then on to the next round. Each round makes the road a little smoother.

An inventory prompt you can use right away

Please help me take stock of this task and judge whether it currently looks more like a dirt road, an asphalt road, or a highway.

Task name: [fill in]
Documents and data sources currently used: [fill in]
Current approach: [fill in]
Deliverable to produce when done: [fill in]

Please work through the following in order:
1. Identify the places where data sources are unclear, versions are a mess, or context is missing.
2. Find the judgments and steps that exist only in my head and haven't been written down yet.
3. Organize the task into inputs, steps, outputs, human confirmation points and acceptance criteria.
4. Suggest the minimum set of documents I need to add or organize.
5. List the parts to hand to the AI in the first round, and the parts to keep handling manually for now.
6. Finally, give me a minimal road-paving checklist I can finish within seven days.

Don't build full automation for me outright. First lay out the current state, the gaps, and the smallest next step clearly.

What this approach solves, and where its limits are

Organizing the environment cuts down on the model's guessing, makes tasks steadier and easier to repeat, and makes it easier to switch tools. It does not guarantee that every output is correct.

Once the road is paved, a human still has to decide the destination, the traffic rules, and when to hit the brakes.

Pave the road first, then bring on the supercar

Stronger models are worth looking forward to. As models improve, the amount of work we can hand off really does grow.

What truly accumulates over the long term is your own documents, knowledge, rules and workflow. While you're still on a dirt road, start by organizing one recurring task. Once the asphalt is laid, add the human confirmation points and acceptance criteria. And once knowledge, processes, permissions and write-back are all connected, let a stronger Agent take over longer tasks.

Extended reading