When the AI Goes Down, What Do You Do?

Read on: My website

Read time: 2 Minutes

I had a great conversation about AI with a friend over the weekend.

Yes, I know, riveting conversation topic: we are in this weird time where the World Cup is over, and Fantasy Football has not started yet, so please bear with me here during these trying times.

We were talking about all sorts of things: work, rugby, travel…the list goes on. On the work front, he mentioned that his company is all in on AI. They are using it for every single application that they can. Constant training and utilization for wherever it can be deployed.

Then he mentioned, “when Claude is not working, all hell breaks loose, and no one remembers how to do their job, talk to a client, or what they are supposed to do next.”

As we continued the conversation discussing this, it unfolded the level of how AI can take over such critical aspects of our daily work, to the point that it seems that some have completely forgotten all their training and education that they have done before AI was embedded into their company.

Was this newsletter forwarded to you?

Dependency can have its pros and cons. It is great that a tool can help you do more with limited staff. But being so dependent on a tool (that has hourly/daily/financial caps) can backfire. Having an AI tool that suddenly hits a limit, the user cannot go any further; there are ramifications on how this plays out if the task or project is not completed.

So…what do you do? What is your team supposed to do?

If you don’t have $100k+ a month to spend on AI, how do you make sure you never get put into a position where you hit a limit, then are unsure where to go? Or Claude just goes down, and you aren’t able to access it?

Let’s dance.

Thinking about hiring globally? Start with an EOR.

The best person for your next role might not live near your office—or even in the same country.

More companies are realizing they don’t need to open entities everywhere just to access global talent. Instead, they’re using EOR to hire internationally faster, stay compliant, and avoid building local infrastructure before they’re ready.

Oyster’s EOR helps companies hire, pay, and support employees in 180+ countries while Oyster handles payroll, compliance, taxes, and local employment requirements.

See how it works

The Pause

This conversation keeps coming up.

At tradeshows. In client meetings. Over coffee between sessions at industry events. Leaders talking about AI adoption with genuine excitement, real results, measurable efficiency gains, and processes that used to require three people now running on one system with minimal oversight.

And then, almost always, somewhere in that same conversation, a quieter moment. A pause. I keep pondering on “We have not really thought about what happens if it goes down.”

Not if the system crashes. Not if the vendor has an outage. What happens if the AI that now runs a critical function fails, and the people who used to run that function manually are no longer there?

As in,

• What happens if you are using AI for your proposal development, and there is a system outage? What do you do?

• What do you do if you are using AI for customer account management growth, and you hit your monthly token or spend limit? Do you remember how to grow your account without AI?

That pause is the most important moment in every one of those conversations. Because what it reveals is not a technology risk. It is a leadership risk that most organizations have been accumulating quietly for years without a single person approving it, designing it, or even noticing it was happening.

What to check out past newsletters? Check them out here:

First Response Wins 

The One NotebookLM Feature That Makes Deep Research Actually Usable 

The Key…in Key Account Management Transitions 

Scenario

A proposal is due in seventy-two hours.

The BD team has been using AI to accelerate every phase of the pursuit. Research synthesis. Competitive positioning. Technical narrative. Past performance summaries. The system has been running well for eight months, and the team has built their entire process around it.

At hour forty-eight, the AI platform goes down. Vendor outage. No estimated restoration time.

The proposal manager pulls up the manual workflow documentation, last updated fourteen months ago, before the AI system was fully integrated into the process. It references a technical writer who left the company four months ago. The compliance review checklist points to a template library that was migrated and restructured when the AI platform was onboarded. Three of the five past performance references in the system were formatted and stored in a proprietary AI output format that the team cannot access or edit without the platform.

The subject matter expert who used to write the technical approach section from scratch now oversees the AI prompting process. She has not written a cold technical narrative in over a year. The junior team members who joined after the AI integration have never done it at all.

At hour sixty, the team is not behind on the proposal. They are starting over.

The AI platform came back online with four hours left before submission. That time they got lucky. But the manual capability that used to be the fallback was gone. Not eliminated by a decision. Replaced by a dependency nobody had formally named, tracked, or tested until the moment the platform went dark.

The proposal was submitted. Barely. And the debrief conversation that followed was not about the content. It was about how close they had come to missing the window entirely because of a risk nobody had put on the register.

Was the proposal compliant? Who knows. The workflow is critically important, as these checks and balances are important to have, especially as the costs for AI tools continue to grow.

Replacement

Every major business continuity framework rests on the same foundational premise. When technology fails, people can step in.

AI breaks that logic in a way that traditional automation never did.

Previous generations of automation assisted human processes. They made experienced people faster, more consistent, and more scalable.

But the knowledge, the judgment, the pattern recognition, the exception handling that came from years of doing the work, that stayed with the people.

When the system failed, the people who understood the underlying process were still there. They could step in, slow things down, run the manual version, and keep the operation functional while the technology was restored.

AI does not assist the process. It replaces the judgment the process required. It takes over the exception handling, the pattern recognition, the decision-making that used to live in people’s heads after years of experience. And when it does that well enough, organizations stop needing the people who held that knowledge. 

Those people get restructured out, reassigned, or simply replaced by someone whose job is to manage the AI system rather than understand the process it replaced.

The knowledge leaves with them. Quietly. Without an alert. Without a threshold being crossed that anyone can see on a dashboard.

Nobody designed it this way. The human fallback was never a formal system. It was a byproduct of how work used to be done. And AI adoption removes it the same way, gradually, invisibly, without anyone realizing what is being lost until the moment they need it.

Continuity

Three things make the vanishing human fallback uniquely dangerous for organizations that have not explicitly addressed it.

1. It is completely invisible.

   There is no alert for institutional knowledge falling below a critical threshold. There is no dashboard metric for subject matter expertise at risk. Organizations do not know they have lost the capability until the moment they try to use it. And by then, the AI is already down, and the clock is already running.

2. It compounds silently.

   Every departure, every restructuring, every training program that gets eliminated because the AI handles it now, makes the gap wider. Competence that is not practiced erodes. The person who knew the manual process three years ago and has spent the last two years overseeing the AI system that replaced it no longer has the same capability they once did. That erosion is cumulative, invisible, and accelerating in every organization that adopted AI aggressively over the last three years.

3. It cannot be recovered quickly.

   You can restore a system backup in hours. You can bring a vendor’s engineering team online in hours. You cannot restore fifteen years of subject matter expertise in hours. When the AI goes down, and the human fallback is gone, the question is not simply how do we recover. It is how long will we operate without a capability we assumed we had, and what does that cost us in customers, contracts, regulatory exposure, and trust that took years to build.

That combination, invisible accumulation, silent compounding, and slow recovery, is a risk profile that most continuity plans were never designed to address. They were built for a world where human knowledge was the constant and technology was the variable. AI inverted that relationship without most organizations updating the plan.

Own It

This is not a question for your IT department. It is not a question for your business continuity manager. It is a question for the leadership team that approved the AI adoption roadmap and owns the operational risk that came with it.

For every critical function that an AI system now runs in your organization, is there a human being who currently has the knowledge, the access, the workflow, and the practiced capability to run that process manually if the system fails today? And tomorrow?

Not theoretically. Not according to an org chart. Actually, right now, today.

The customer doesn’t care if your AI fails; they care that you support them and answer their needs.

If the answer to that question is unclear for any critical function, you have a gap that your continuity plan does not see and your risk register does not reflect. And that gap is widening every time someone with institutional knowledge leaves, every time a training program gets cut, and every time a new hire is brought in to manage the AI rather than understand what it replaced.

The organizations that will navigate the first serious wave of AI operational failures are not the ones with the most sophisticated systems. They are the ones whose leadership asked this question before the system went down rather than after.

Identify your three most critical AI-dependent functions. The ones where a failure would immediately affect customers, contracts, or regulatory obligations.

For each one, answer a single question honestly. If this system failed today and stayed down for forty-eight hours, who specifically would run this process manually, and when did they last actually do it?

Does your team remember how to do this without AI?

If you cannot answer that question cleanly for any of the three, you have found your continuity gap. Not a theoretical one. A real one that exists right now inside an organization that has been assuming it was protected.

Fix the assumption before the system tests it for you.

Because when the AI goes down, the question of who steps in will get answered one way or another.

The only variable is whether you answered it first.

See you next week.

Whenever You’re Ready, Here are 4 Ways I Can Help You:

1. Unlocking Hidden Potential – Reconnecting with Past Clients for Explosive Growth – Check out my free eBook on how you can find hidden gems in your past clients and help you crush your sales goals.

2. AI for Business Development – Download our free eBook on how you can effectively leverage AI prompts to your advantage. From properly setting up your preferred AI tool, to how to shape your prompts, save time, and get the outputs you are looking for.

3. Sales Resources at Your Fingertips – From tools, tips, demos, and how-tos, check out our Pages and content that can provide you with additional support, whether it be social selling, account management, or something else.

4. Cribworks Advisor Program – Want more than just resources? Reach out to me and see if our Advisor Program can help you scale your business.

Pat Miller’s – Fridays Off Newsletter

You didn’t start your small business to be owned by it. Fridays Off gives you one weekly step to reclaim your time, increase profit, and build a business that gives you your life back.

Subscribe

Superhuman AI

Keep up with the latest AI news, trends, and tools in just 3 minutes a day. Join 1,500,000+ professionals.

Subscribe