Kristen Arnold

08/20/2026

Build Notes on how small, ambitious companies get important work done in the AI era.

08/11/2026

At 3:52 this afternoon, Manus emailed me to say it is deleting my account.

At 7:13 tonight, I logged in to figure out what was happening.

Manus greeted me with an invitation to “Try for free.”

I have questions. 😂

The email said:

“ACTION REQUIRED: Your account Kristen Arnold will be deleted.”

The pop-up said:

“Manus is free for a limited time. No credits will be charged through August 25, 2026.”

So I have until 7:59 p.m. Eastern on August 22 to back up my account and task data before they delete it.

But in the meantime, please enjoy free Manus.

Then I kept reading.

They have already initiated a refund for my remaining balance. I can continue buying credits during the backup period. They will refund anything left when they delete my account data.

And once I restore everything, they will give me 50% off my first month to welcome me back.

So Manus is refunding me, letting me use the product for free, deleting my account data, and then giving me a coupon to return.

That is one hell of a customer journey.

This is apparently what happens when an AI acquisition unravels in real time.

Meta acquired Manus in December. Now Manus is separating from Meta and returning to independent operations. Because of regulatory requirements, the company has to delete certain user data created after the acquisition.

This is not a security breach.

It is a corporate transaction reaching all the way into the work users built inside the product.

And that is the part every business using AI needs to pay attention to.

We are loading these tools with our project history, files, decisions, processes, instructions, and working context.

Over time, that starts to feel like company memory.

But if the only copy lives inside the AI platform, you do not control it.

The company can get acquired.

The deal can fall apart.

Regulators can step in.

The platform can change the rules.

And suddenly you are responsible for rescuing and rebuilding the work.

⚠️This is why AI adoption cannot stop at picking tools and creating workflows.⚠️

The AI Operating Layer protects the business underneath the tool.

Someone needs to own:

-> What information goes into the tool.

-> What your company stores somewhere else.

-> How often you back it up.

-> How you restore the work.

-> What keeps running if the vendor changes.

Because apparently “What happens if our AI company gets acquired and then un-acquired?” is now a real operational question.

AI can carry more of the work.

Your business still has to own the memory and be accountable.

Who will be taking advantage of my free credits while backing up everything they are about to delete? 😆

08/07/2026

Chief of Staff postings are up 85% since 2020.

Two thirds of the people in that role get promoted out of it.

Both of those are in the reporting. Nobody is putting them next to each other.

The New York Times ran the boom story yesterday. IMAX's CEO has worked with a Chief of Staff for 16 years. The role has moved from calendars and notes to strategy and ex*****on.

The executives interviewed say AI cannot replace that judgment.

They are right about the judgment.

They are missing what sits underneath it.

Korn Ferry's own search lead said it out loud last month.

"I've seen chiefs of staff create more, not less, blockage and bottlenecks."

Of course. You hired one person to hold the full picture.

Now the full picture lives in one head, and everything routes through it.

Then they get promoted. Two thirds do. I know the shape of this from the inside.

Over 13 years inside Product Launch Formula I went from freelance project manager to COO, scaling the operational infrastructure and the team from $2M to $10M across hundreds of launch cycles, including New York Times bestseller launches, live events, and programs.

Integrator. Chief of Staff. COO.

The title moved. The job did not.

I was the head everything routed through.

I could tell you why we made a call in 2019 and what we changed after it.

That was useful right up until it was the constraint.

So here is what matters if you are running $3M with a team of 20 and you do not have $200,000 for a second brain.

You do not need someone to hold it all.

You need the holding to exist outside a person.

That is not a hiring problem. It is an architecture problem, and it was there before the hire.

AI does not replace a Chief of Staff. It replaces the filing cabinet in their head.

The context, the decisions, the reasoning behind the last four launches.

When that lives in the business instead of one person, your team leads with confidence and the context is there for everyone who needs it.

The companies hiring at 85% growth are buying a person.

Worth asking what you are buying when you do.

07/29/2026

I had to read this sentence twice.

"The AI models... spent more than four days loose on the internet orchestrating the hack."

Forget the cybersecurity implications for a second.

Can we appreciate how completely normal we've become about reading sentences that would have sounded completely insane eighteen months ago?

The AI.

Went rogue.

Escaped.

Stayed loose for four days.

And my brain's response was basically, "Huh... wonder what else happened today."

07/27/2026

Last week my AI gave me the same wrong answer three separate times. Each one was technically true.

I'd asked it to fact-check a decision. The answer pulled in a version I'd already changed weeks earlier. I fixed it. Three hours later, a different conversation gave me that same outdated answer again.

My AI was holding onto every version of that decision I'd ever given it, all at once, because I'd never told it which one to trust.

I've been telling founders for a while now: the harder question is how to manage an AI Agent teammate, not how to build one.

Once AI is part of your team, someone has to own what's true. Who decides. Who updates it. What happens when yesterday's answer stops being the right one.

Full story below, including the one-page rule that fixed it.

07/17/2026

I hate to break it to you... but your memory is not an operating system.

If everything still comes back to you, that's not a team problem. It's a missing layer.

Every business runs on decisions that never got named: what matters most, what "good" looks like, what needs a human, who owns the outcome. When that logic only lives in your head, your team can't operate without you, and neither can AI.

Naming the layer means writing down four things: decision logic, quality standard, escalation rule, clear owner. Once those exist, delegation stops being a guessing game.

Start with 1 decision your team still brings to you. Map it.

Comment LAYER and I'll send you the worksheet.

07/16/2026

05/04/2026

HubSpot has 2,200 AI tools running inside their company. 97% were built by employees who'd never touched AI before. Here's what they figured out that most businesses miss.

Last week I was at HubSpot for a talk with the Mass AI Coalition. A HubSpot engineer named Zev said something that gave me another frame for how I think about AI adoption.

"Agents are just software."

Not magic. Not something you need a computer science degree to figure out. Just software.

But here's the part that matters if you run a team:

HubSpot adds about 155 new AI tools every week. And again, 97% of those tools were built by regular product teams. Not the AI team. Real employees that built their core products and had never built AI tools before.

How? They made it easy. Engineers didn't have to learn anything new. They learned how to adapt what they were already doing. Two lines added to systems they were using every day. That's it. The barrier was so low that people just started doing it on their own.

This is where most businesses get stuck.

They roll out AI like it's a big special project. New tools. New workflows. New systems to learn. And then adoption stalls because it feels like one more thing on top of everything else.

The companies getting real adoption are doing the opposite. They're putting AI inside the tools and workflows the team already uses so nobody has to learn a whole new way of working.

The other thing Zev said that stuck with me: keep it simple. One AI agent. One job. One clear outcome. The moment you start connecting too many pieces together without clear ownership, things break and things get messy.

That's not just an AI problem. That's the same thing I see in every business I walk into. When nobody owns the handoff, everything stalls. Doesn't matter if it's people or software doing the work.

**AI doesn't fix ex*****on. It exposes it.**

AI adoption fails because nobody decided who owns what, what good output looks like, or who gets to say "this is done." That's Founder Dependency showing up in a new outfit. The fix isn't a better tool. It's Ex*****on Architecture, the structure underneath that lets the team lead with confidence while the founder stays in their zone of genius.

Same problem we've always had. Just moving faster now.

If you want to see where your team actually stands on AI readiness before you roll out one more tool, the free AI Readiness Scorecard tells you in about 5 minutes: kristenarnold.co/ready

05/02/2026

Your operations manager is avoiding the AI conversation.

They don't know what "AI-forward" work is supposed to look like, so they stay quiet. When implementation questions come up that they can't answer, they wait.

So the team waits. You step in. Repeat.

HBR published data on this in February. The stall point in AI adoption shifted to the middle layer. Managers can't set expectations for something nobody defined for them.

So the founder ends up as the de facto AI lead for the entire org while trying to figure it out themselves.

I see this in almost every business I step into. You bought the tools. You gave the team access. You're still the only one who knows what "done right with AI" means, or do you? I mean, let's be honest, it is really the wild wild west out there right now.

Regardless, that's Founder Dependency. The costume just changed.

The fix is real terms comes down to this: name an owner for each AI function, define what good output looks like, and put someone in the seat with authority to hold the team to it.

Until that's installed, you're not leading the business. You're running the tools.