Models are like supercars, knowledge base and workflow are the road under your feet

What is this article talking about?

When a new generation of strong models such as Claude Fable and GPT-5.6 Sol appear, many people’s first question is: Do they need to upgrade now? Will it be too late? I'll look at one other thing first. Where have your knowledge base, workflow and knowledge flow been organized? The same supercar is completely different when driven on different roads. Who is

Fable and GPT-5.6 Sol both remind the same thing

Anthropic’s Claude Fable and OpenAI’s GPT-5.6 Sol have different capability routes and delivery methods. What the two have in common is that the model can begin to handle longer and more complex knowledge work, research and tool operations.

As model capabilities continue to improve, the files, rules and workflows originally prepared by users will also more directly affect the final experience of use. Similarly, if you get a powerful model, some people can use it for long-term tasks, while others can still only use it for a single question and answer.

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

A supercar, placed on three types of roads

When automobiles were first invented, their speed and reliability did not necessarily surpass horse-drawn carriages immediately. Some people only see the comparison in front of them, while others begin to see another possibility: the car will continue to progress and the road will change accordingly.

When the road is still mud and gravel, the difference in feeling between a slower car and a faster car is limited. When the road surface turns into asphalt, the stability and speed of the car begin to be seen. Only when highways appeared did supercars really have room to perform. Strong models such as

Niba Road: Information is scattered, and the process is memorized by people

Mud Road can still use AI. General Q&A, single revision, and summary editing can still be done quickly. The problem occurs when the tasks start to become longer, the data starts to increase, and the judgment starts to become complicated.

• There are several versions of the same project, and I don’t know which one is the latest.
• The working method only exists in my own head, and there are no written steps.
• AI needs to be asked about the background, audience, format and restrictions every time it starts work.
• The data is scattered in LINE, Email, cloud drive, chat history and various note-taking tools.
• There are no acceptance conditions after completion, and the new experience is not written back to the original file.

At this time, the model can only guess the path on the loose ground. As the model becomes stronger, it can indeed guess more correctly. Without data, rules, and goals, it still doesn't have enough to work with.

Asphalt Road: Documents centralized, workflow begins to solidify

The key to the asphalt road is to prevent the AI from understanding you from scratch every time.

At this level, articles, briefings, meeting arrangements and research reports can begin to form a stable workflow. After the model is upgraded, it is easier to feel the difference because the tasks already have a clear route.

Highway: Knowledge, rules, authority and acceptance are connected

After these structures are connected, AI has the opportunity to move from answering questions to taking over a piece of work.

My files are my system

I started using Obsidian in 2024, and later organized it all the way to driving engineering and loop engineering. This process has made me more and more certain of one thing: tools can be replaced, and my own files must continue to accumulate.

Moving costs are usually lower if the files are in common formats such as Markdown, plain text, CSV, etc. Change a set of AI, change an Agent, change a note-taking tool, and the originally organized content can still be used.

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

This is what I call a cycle project: input, execution, acceptance, correction, recording, and then entering the next round. Each round makes the road a little smoother.

An inventory prompt that can be used directly

Please help me take stock of this work and determine whether it is more like a dirt road, an asphalt road, or a highway.

Job title: [fill in]
Documents and data sources currently used: [Fill in]
Current practice: [fill in]
Results to be delivered upon completion: [Fill in]

Please complete in order:
1. Identify areas where data sources are unknown, versions are confusing, and background is lacking.
2. Find judgments and steps that only exist in my mind and have not yet been written down.
3. Organize the work into inputs, steps, outputs, manual confirmation points, and acceptance methods.
4. Suggest which files need to be added or organized at least.
5. List the parts that will be handed over to AI in the first round, and the parts that will be temporarily retained for manual processing.
6. Finally give me a minimum paving list that can be completed in seven days.

Don't just do full automation for me. First, clearly explain the current situation, gap and minimum next step.

What can this method solve and what are the boundaries?

Decluttering the environment can reduce model guessing, make tasks more stable, easier to repeat, and easier to change tools. It does not guarantee correct output every time.

After the road is paved, people still have to decide their destination, traffic rules and when to brake.

Pave the road first, then get on the supercar

Stronger models are worth looking forward to. As the model improves, the amount of work we can hand over will indeed increase.

What can really be accumulated over the long term are your own files, knowledge, rules and workflow. When the road is still muddy, sort out a repetitive task first. When the asphalt road is paved, manual confirmation points and acceptance methods are added. When knowledge, processes, permissions and write-back are all connected, a stronger Agent can take over longer tasks.

Extended reading