A note from Max and Clare. We recently reconnected to compare notes on AI. One conversation turned into a three-part series about how we see it, what it means to build on a stack we don't own and what we choose to do with the capacity it creates.
We don't agree on everything. That's what made the conversation worth sharing. We hope you enjoy the series.
Part II of Counterweight. Start with Part I → · Read Part III →
Clare: The Bag She Packed
A client wanted me to run on their hardware. Initially I pushed back. My operating system had been built over months inside one toolchain. Context files, agents, skills, workflows. I didn't know how to disentangle any of it for one engagement.
But the client's CISO wouldn't budge. So I spent half a day building a standalone repo with enough context about me and OxerLine to bring to their virtual machine. And it worked, for the most part. I still pulled a few more things in mid-engagement, mostly my design system. It was CISO compliant: a one-way door where my context could go in, but their company context couldn't come out. All of my lessons got captured and built into a skill for next time.
The benefit I hadn't originally priced? I ran the engagement on their tokens, not mine. All 18 million of them.
What I'd Actually Proven
The exercise slowed me down at the start. It also gave me something I hadn't been building toward. I could meet a compliance-sensitive client where they needed me without rebuilding from scratch. My core context wasn't welded to one toolchain or one laptop.
I felt good about that for about a day.
Then I looked at what was still sitting in my main repo. I'd proven I could move the bag. But I hadn't proven I'd packed light.
How I Pack
In my physical life, I have strong opinions about packing light. I'm a carry-on only kind of person. Shoes get debated. Toiletries get consolidated. I know what fits in a personal item and I have feelings about it.
But the repo with my company operating system? It's heavy and I shrug.
It's not a secret. I know I could be more tidy and retire unused context files and skills. There's very little cost for the weight right now, so why spend the time optimizing? To quote Blaise Pascal, "I only made this longer because I didn't have the time to make it shorter."[1]
This subsidy may go away eventually. Anthropic and OpenAI are burning cash toward IPO. Google and the rest are subsidizing access to hold ground. But for now, the rest of us experiment on someone else's dime. I've shipped things I couldn't ship a year ago. Engagements run faster. Ideas that used to die in the someday pile get a prototype before dinner.
The subsidy doesn't just lower the cost of good work. It lowers the cost of sloppy work. You stop asking whether something needs to exist.
Even though I know the failure mode, I'm doing a version of it inside my own operating system, adding context files and half-baked workflows because the marginal cost of one more thing is basically zero.
Portability Is Not Efficiency
I keep wondering whether efficiency will matter.
My husband makes video games. A few years ago, we were chatting about the evolution of graphics cards. When the cards were weak, teams optimized everything to ship something that looked good. Down to the machine code. The 1s and 0s. But as the hardware got better, developers could be lazier. The console forgave them.
AI feels like it is the reverse: most of us have started off as lazy developers. We haven't learned the discipline of optimization.
I don't know which way this goes. Does the subsidy suddenly dry up? Or do we get a graphics-card moment where we're forgiven for the bloat we've accumulated while experimentation was free?
I'm trying to learn lean anyway. I don't want to bet wrong.
What Might Close the Window First
Compute might decide the bet before any of that resolves. It's the processing power behind the API call. Cloud capacity. Inference.
It's not something many people think about, but it's not unlimited. A friend at a boutique UK consulting firm projected they'd run out of cloud compute. Not someday. In the next 12 months or so. And when it's a scarce resource, it will go to big players. Not small firms. If you're not on the list, you're not on the list.
Max said what I was already thinking: we're both women, we're both solopreneurs. We might not get any compute. That's not portability and it's not efficiency. That's access. And it might hit before we find out whether the subsidy or the models blink first.
Max: The Bag I Never Packed
Clare ran her fire drill on purpose. Mine ran me.
The Folders Were All There
A few weeks ago my Mac decided, on its own, that my cabinet didn't need to live on my desktop anymore. The sync settings I never thought about quietly started moving everything to the cloud, and what was left behind looked like my files but wasn't. The folders were all there. The contents were gone. My whole operation, the bots I spent three months training, the memory files, the voice rules, all of it suddenly unreachable by the tools that run my business every day.
I spent a day and a half untangling it. And here's the part that humbled me: I wasn't really the one fixing it. The AI was fixing itself while I sat there approving things (chaperoning, basically). That's when the question landed. If Claude disappeared tomorrow, what do I actually have? At that moment, the honest answer was nothing. No backup. No copy of the thing I'd describe as my company's operating system.
Sync Settings Versus a Continuity Strategy
Nothing broke, thankfully. But I came out the other side a different operator. Everything now lives in three places: local disk, a Google Drive mirror, and a frozen snapshot that gets emailed to me. And the next move is getting the whole thing onto a Mac mini that sits on my desk and answers to no one's sync settings but mine. Not because I think Claude is going anywhere tomorrow, but because I finally understood the difference between two things I'd been treating as the same: sync settings and a continuity strategy. One is a checkbox. The other is knowing, on paper, what your business needs to still exist next week if any single vendor vanishes.
Clare packed her bag by choice and found out she could. I found out I'd never packed one. Her 18 million tokens on someone else's tab and my day and a half of chaperone duty are the same lesson wearing different clothes: this subsidized window is the cheapest tuition we will ever get. Use it to find out what actually needs to persist, while the cost of finding out is a scare instead of a complete and total loss.
About the Authors
Clare Hawthorne: I founded OxerLine Advisory to fix friction at the seam between technology and the business. If this is something you're facing, let's chat.
Max Tuchman: Runs Dark Horse Ops, private AI tutoring for executives and founders. She's a founder (Caribu, acquired by Mattel), former GM at Mattel, White House Fellow and high-school teacher. She now works with leaders 1:1 to move from chatting with AI to building real capacity with it.
Sources
[1] Blaise Pascal, Lettres Provinciales (Provincial Letters), 1657: "Je n'ai fait celle-ci plus longue que parceque je n'ai pas eu le loisir de la faire plus courte." Often misattributed to other authors as "If I had more time, I would have written a shorter letter."
A note on how this got written: AI assisted Clare and me in capturing our conversation and organizing our thoughts into three posts. The stories, the opinions, and the scars are all ours, but we work with very good robot assistants who help us get them out of our heads and onto the page.