What I’m thinking about: Claude Code moved an entire company for me in about a week…and the one page it couldn’t get right is the clearest picture I have of what I actually get paid for now.
Seventeen of the ~20 answers on my new FAQ page had to be rewritten by hand.
Last week I shared that enterprise AI implementation, under Serious AI Partners, is now my full-time work (and this newsletter is now Serious AI News…still every Monday). It had been ~7 days since that decision when I recorded this week’s podcast, and nearly all of them went into moving the ‘piping’ from Serious Land Capital (SLC) to Serious AI.
What a week with Claude Code replaced
A week ago, I had an LLC and a bare static homepage hanging off the SLC site, and that was about it. Everything else had to move: a full website, DNS and email deliverability, SEO and GEO (getting cited inside AI answers, as opposed to ranked on Google), the podcast/newsletter and their publishing routines, public-facing social accounts, and the software stack underneath it all.
In past businesses this was a multi-week slog (for SLC we spent thousands of dollars on website design, SEO, and routine technical work), and I used to hire a Fiverr contractor just to set up DNS for any new email domain. This time nearly all of it ran through Claude Code on Opus 5, at ~10% of my weekly Claude Max allowance, despite near-constant usage…just stunning (and more technically optimized than prior work).
Go take a look at the result before reading on.
It also helped that I wasn’t under cash pressure to land a client inside the first week (though initial conversations have been going even better than I’d hoped, reinforcing the business move). That bought the room to work ON the company before running it (similar to how we started SLC, which oiled up the machine for a roaring start).
“Where are the spoons?”
Anyone who has moved houses/apartments knows the feeling. Being in a new place is exciting, but for the first few days you still can’t find the utensils, the drive to the gym takes actual thought, and all of that ‘re-wiring’ costs mental capacity, preventing forward progress.
That is the price of starting something new, and it’s the part entrepreneurs (myself very much included…as I’ve learned more than once) underestimate when they open a second or third (…or eighth) thing. Distracting priorities are almost always the largest threat to your potential, so with full conviction in the full-time business switch, I chose to take the pain all at once.
Claude Code was the moving crew. It carried every box across town in a fraction of the usual time, throwing out some items that had outlived their use, and purchasing new furniture along the way…but it still fell to me to decide on HOW I wanted my new home to look and feel (to create the ‘feng shui’, if you will, and more so for my guests than for myself).
^ I’m especially proud of this analogy…hopefully it resonated.
Remember, there are no rules in business (except to not run out of cash, or lose the ability to generate cash), so when you have an opportunity to set the scene precisely to your specifications, do so in a way that will produce the most long-term satisfaction (and ideally maximize value-add to your customers and partners).
Where the moving crew stopped being useful
For any business with a digital footprint (arguably every business should have one), its website is still the most important marketing asset IMO, primarily because it’s the only place that can serve as an ungated source of truth for your brand. We spent a ton of time on SLC’s website during our launch (and throughout the past few years), and Serious AI Partners was no different.
During the build-out, the FAQ page is where I felt the limitations of Claude Code the most. I pointed Claude at a handful of sites for inspiration, then had it run the draft through my voice library and everything we’d defined about the new brand so far (it has years of my writing and reference material to draw from).
The first pass produced 7 answers and ZERO survived intact. By the time the page had grown to the ~20 questions it needed, maybe 2 or 3 of the model’s answers shipped as written.
For example, on the initial question about whether we train client teams in AI, the draft said “No” to something we do.
And it presented the ‘Check’ (from Sharran Srivatsaa’s four-column framework: Decide, Build, Check, Run) as house language with no credit, and it assumed the reader would know what that four-column framework was, instead of including a brief primer (the full FAQ page is here if you’d like to see what survived).
The Build took minutes, and the Check took a week. Most of the website and social profile AI drafts (worth reinforcing, AI drafts built upon tons of rules and guidance from my own existing writing style and positioning) didn’t read as glaringly wrong. They read as smart, polished and finished, which is the danger, because you stop checking at the exact moment the tool sounds most sure of itself (researchers call this automation bias, and it gets worse as the tools get better).
Enterprise buyers are among the most sophisticated customers in the world, and a newer entrant into AI implementation (like me) can rule itself out with one careless answer.
The $250M version of the same gap
I’ve been following a company called Polsia closely. Solo founder Ben Cera raised $30M at a $250M valuation in May with zero employees, on AI agents that promise to build and run a company for you (the dream almost every entrepreneur has had, and perhaps every human), and by his own account was approaching a $10M annual run rate.
On the surface, an incredible data point, and representative of the leverage AI can give when properly wielded…but then there are the customers. When I checked Polsia’s Trustpilot page about a month ago it sat at 1.8 out of 5 across 79 reviews, 74% of them one star, with complaints about tasks marked complete that never deployed and credits consumed by failed runs.
On a re-check, as of this weekend, Trustpilot reads 3.0 across 210 reviews (more new reviews in the last month than in the prior six), on a profile the company claimed in August.
One reviewer on Sept 17 complained that they “still haven’t received the 10 credits from leaving my last review,” on a platform whose rules bar companies from offering incentives. That’s one reviewer’s statement, and from the outside I can’t tell whether the product improved that fast or the mix of people being asked changed. A customer can’t tell either.
Regardless, both things are true: the ceiling on a one-person company has significantly risen, and the reviews (on Polsia, at least) say the Check is where there’s still room to carve out a moat.
My read, after a week of running the best publicly available models, and still needing to perform a thorough Check myself on nearly every paragraph of new website and branding copy, is that a business can’t simply be handed to AI agents on autopilot without review…least of all by customers who haven’t yet had the reps to grade the output. They bought what the market is selling, and the market is selling the Build while skipping the Check.
The wider data rhymes. In McKinsey’s State of AI in 2026 survey, 32% of organizations skipped at least one software purchase because coding agents could build it in-house, and last week alone (per SiliconANGLE’s enterprise roundup ) three vendors shipped kill switches or rollbacks for AI agents while a startup raised $40M to certify them.
Building got cheap, and the market is starting to pay for the Check. My one caution is that a generic check that ships with the tool is the same mistake one level up…verification has to come out of your own (or your team’s) repetitions, and the things that have already cost you money.
Brand first, then go hog wild
There’s a fair argument for chasing customers first and sorting out the brand later. I’ve come to prefer the reverse (as noted, we did the same at SLC), because building the About page and the FAQ forced me to come up with answers I’d otherwise be working out live in front of a potential client. As Hormozi puts it, “an ounce of pre-work is worth a pound of post.”
With the picture set, the marketing side can run hard. When I finish recording a podcast now, a Claude Code routine cuts it up, proposes headlines based on what has performed for us, writes the Spotify and YouTube descriptions, hands the thumbnail prompt to ChatGPT (best-in-class image gen at the moment), takes back the output, asks for my final approval…and then publishes across all platforms.
Once a routine has passed enough of those approvals, it won’t need me to post at all. Same posture as always: validate on the first passes, then trust and move.
I’ve yet to encounter a business (mine included) that is 100% AI optimized, so the work is choosing where the next Check belongs. I’m very proud of how seriousai.partners has turned out (in just a week!), so click through and judge for yourself whether it reads like me or like AI slop. You’re welcome to reverse-engineer any of it (including the backend metadata), my updated LinkedIn profile included…as my gift to you all.
The moving crew gets faster every month, and I’ll keep hiring it for everything it can carry. Knowing that the kitchen table should be placed ‘just so’ stays my job, and it’s the part clients are paying for. Serious means underwritten.
=====
If you run a ~$5M+ business and AI output is leaving your building faster than anyone can check it, the work starts with diagnosing the constraint that’s gating you, then building with verification sitting where a human still owns the outcome. It’s how you become a business whose AI work still holds on the hundredth (or hundred-thousandth) run. Everything about how that works lives at seriousai.partners.