Field Notes
The easiest mistake is thinking Blockbuster lost because Netflix had better technology.
That is too clean.
Blockbuster had stores. Customers. Brand recognition. Distribution. Cash flow. A physical footprint all over the country. People knew the name. People knew the routine. Friday night, drive to the store, walk the aisles, grab the movie, pick up candy, bring it home.
Then the ground moved.
The customer did not want the better version of a late-fee business forever. They wanted the movie without the friction.
Netflix was not only a website. It was a different operating model.
That is the part I keep coming back to with AI.
A lot of companies are trying to become the AI version of Blockbuster.
They have the internal memo, the pilot, the Slack channel, and the executive saying, "We are looking at AI." A few seats of a tool get bought. A few people experiment. Someone may even write a policy document that tells everyone what they are allowed to do.
From the outside, that can look like movement.
From the inside, the work often has not changed.
The same person still has to remember the process. The real operating system still sits inside a spreadsheet. Teams still ask who has the latest version. Managers still review work with no trace of what changed. Data still gets trapped behind a screen with no clean way for another tool to read, write, or act.
That is the Blockbuster Test.
When the market changes, is the company changing the way work actually gets done, or is it decorating the old system with new tools?
This matters if you run the company.
It also matters if you work inside one.
Because the employee question is getting sharper:
Am I working inside a company that is adapting, or am I working inside a company that is slowly becoming the place everyone talks about later?
In Welcome to the New World, I wrote that we are moving from the old world into the new one. This is one of the first practical tests of that claim.
The old world treated software as the system of record and humans as the connective tissue.
The new world treats workflows as the asset. Software, agents, automations, prompts, models, dashboards, and people all have to plug into that workflow.
If the workflow only exists because five people know how to click through it, the company has not built the new operating layer yet.
It has old-world process with new-world cosmetics.
Playbook
Here is how I would run the Blockbuster Test on a company.
Start with the tools, but do not stop there.
A serious company is not simply asking, "Do we have ChatGPT?"
It is asking:
Can our tools be used by people and agents?
Can the data move?
Can the workflow be triggered?
Can actions be logged?
Can the output be reviewed?
Can permissions be scoped?
Can the process run the same way when the person who normally handles it is out for a week?
That is why The Agentic Tool Test matters. A tool that works only as a pretty screen for a human may be fine for the old workflow. It becomes a constraint when the company needs agents, scripts, APIs, MCP servers, CLI access, audit logs, exports, and permissioned tool use.
The next test is documentation.
If a process lives in someone's head, the company is fragile.
If it lives in a recording that no one watches, it is still fragile.
If it lives in a 42-page SOP that is already stale, it is only pretending to be stable.
The question is simple: could a smart person or a well-scoped agent look at the workflow and know what to do next?
That requires a few boring things:
- the trigger;
- the inputs;
- the decision points;
- the tools;
- the permissions;
- the review gate;
- the definition of done;
- the receipt that proves the work happened.
There is no magic in that list. That is why it is so important.
AI exposes whether the company ever understood its own work.
The next test is training.
I do not mean a one-hour "how to prompt" lunch-and-learn.
I mean employees can answer basic operating questions:
When should I use a chatbot?
When should I use a custom GPT or project?
When does this need an automation?
When does this need an agent?
When should I keep a human in the loop?
When is the risk too high to let the system act without review?
When should I stop hacking around the process and ask the company to redesign the workflow?
That last one is the real culture test.
Some companies will punish the person who finds the broken workflow.
The better companies will make that person valuable.
They will say, "Where is the work getting stuck? What are you still doing manually? Which report does no one trust? Which system makes you copy the same information three times?"
That is how the company learns.
The next test is review gates.
AI inside a business cannot mean "let the model do whatever it wants." That is lazy and dangerous.
It also cannot mean "ban everything until legal feels comfortable." That is how cautious companies become slow companies.
The middle path is better:
Let the system draft.
Use it to gather.
Ask it to compare.
Have it prepare the work.
Give it room to flag what needs attention.
Then decide which actions need approval, which actions can run automatically, and which actions should never be delegated.
This is why What Is an Agent, Actually? matters. An agent is useful because it can pursue a goal with tools and context. In a company, that power needs boundaries.
The goal is not chaos.
The goal is controlled throughput.
More work gets prepared, routed, checked, and finished with receipts.
The last test is whether leadership changes the process.
This is the part everyone wants to skip.
Buying tools is easy.
Changing how work moves through the company is harder.
If leadership talks about AI but never changes any workflow, the company is still browsing the aisles at Blockbuster.
If leadership asks every department to identify the manual workflows, document the repeatable ones, choose tools that agents can touch, design review gates, and measure the hours or quality improvements, then the company is actually moving.
That is the difference.
AI adoption is visible in the work, not the announcement.
Orientation
I would use this test in two directions.
If you run a company, use it as an audit.
Pick one workflow this week. Not the whole company. One workflow.
Ask:
- Is it documented?
- Can someone else run it?
- Can an AI system help prepare any part of it?
- Does the tool stack allow that help?
- Where does a human need to review?
- What receipt proves the work happened?
If you work inside a company, use it as a signal.
Watch what leaders reward.
Do they reward people who protect the old process?
Or do they reward people who find the manual work, name the risk, and help rebuild the operating layer?
That answer will tell you a lot.
Some companies are going to adapt quietly before the market forces them to do it loudly.
Some companies are going to keep looking responsible until the customer, employee, or competitor has already moved on.
That is why the Blockbuster story still matters.
The lesson is not "technology wins."
The lesson is that the future usually shows up first as an operating model people can dismiss.
Then one day the customer stops driving to the store.
Next up: Don't Single-Source Your Agent.
Because even if your company passes the Blockbuster Test, there is another risk hiding underneath the stack: what happens when the tool, model, platform, or workflow layer you built around changes the cost, structure, shape, or rules of the work?
That is the AI sovereignty question. Arc 1 was about refusing to marry one model. Arc 2 has been about building an agent-ready stack. Post 019 brings those together before we move into implementation.
Comment below with one sign that tells you a company is moving fast enough, or one sign that tells you it is still pretending.
— Brian