Status: Build 01 is HubSpot Sales Hub Professional.
It started on October 20 and has until January 19, 2027. 78 days left.
Each quarter I pick a piece of software that businesses pay for and rebuild it with AI coding tools. Each build gets 13 weeks to reach a stated goal, and the goal is never "it looks like the real thing". It's real use: teams other than mine running their work on the imitation instead of the tool they pay for.
I'm not sure there is a CRM I haven't tried, yet. Honestly, I don't have many faults with most of them, and they're all more or less the same to me. But it's clear that for a lot of sales teams they don't check enough boxes, and the ones that come close keep getting more expensive. It's possible it's impossible to ever quench the insatiable thirst for UX improvements that sales teams desire from a CRM. Hence, why we find ourselves here with a CRM as our first target.
I've done these types of builds in private for... well... most of my life. I've done them for other people as an employee, they've made a lot of money, and then the world is left to wonder about the secrets of how it was done. And I keep watching people speculate on whether something can or can't be done, while sometimes knowing the exact answer.
In a world where almost nothing is taken at face value, and even facts are disputed immediately, and violently by some, without any evidence to the contrary, no one is going to believe the answer to this type of question without building in public with receipts.
Also, honestly, I don't think every person needs to pay the prices they do for some of these tools. I think it makes sense for some and not for others. People should better understand those options and have more choices.
Despite this being the ‘What this is’ section, I’ll also take a moment to state what this is not. This newsletter is categorically not an indictment on Saas as a category or the prognostication of its impending doom; I think that’s silly. I constantly find Saas companies where their current product, price, roadmap, and speed of execution leave them firmly outside the realm of consideration for a project like this.
But, be assured, that if they rest on their laurels and fall asleep at the wheel, I’ll be more than happy to revisit that conversation.
I turned 42 the day before I started this build. (Wait. Maybe this is a mid-life crisis. Can't be. Never mind. But. Could it be? No. Anyways.) Despite decades of tinkering with technology, I would never claim to be an engineer or coder. I've spent most of my career where technology meets entertainment, and I’m intimately familiar with HubSpot as a product and company, though I don’t have a current subscription.
With the AI, I'm mostly captain of a ship, but I'm not the helmsman, nor crew. I oversee, I guide, I make decisions, I approve, and I make the tough calls, but the AI is executing.
More on both, the person and the method, in later issues. For now, the toolset:
Claude Code on Claude Opus 5.5, on my personal Claude plan. I was on Max 5x ($100 a month), which I already paid for, and on day two I had to upgrade to Max 20x ($200 a month) to keep building. The extra $100 a month is counted as real money; the rest is named here, not added to the build's costs. Its tokens are counted.
OpenCode with open models (DeepSeek and GLM through OpenRouter, paid per call), for simpler work and to keep going when I hit Claude's usage limit. I also use OpenRouter models to perform pentests on the build as Claude throws a tantrum if you even mention security.
The app: Next.js and TypeScript, Postgres on Supabase, hosted on Vercel. I already pay for Supabase Pro and Vercel Pro across my projects, so they're named, not added. The extra $10 a month for this build's Supabase project is counted. Where something hard is already done well, like syncing a Gmail inbox or sending email, I use the existing service (Nylas, Resend) and count its cost. Both are on free plans so far, but they definitely will have to upgrade to paid plans soon.
Cloud test computers, rented by the minute, run the full test suite and the click-through checks on every change. They're paid per use, so they're counted.
Open source, with rules. Every library is listed with its license; nothing under GPL or AGPL. The code is Apache-2.0 and opens to everyone only if a build graduates.
Every issue has a burn rate with three numbers:
Hours are mine. The time the AI spends working is a footnote, not a headline.
Tokens are new tokens used by the coding tools, counted from their own logs. Not "what this would cost at list price".
Real money is only what actually hits a card for the build: hosting, services, API bills. Subscriptions I already pay for are named, not added.
The rules are on the site so you can hold me to them (duped.dev/rules): one quarter per build, every hour logged, every hand named, graduates get maintained, and failures get published with the same detail as wins.
What to expect:
A free issue every Tuesday, starting tomorrow: a one-line status, one main piece (a build update, or field notes on a quiet week; no padding), the burn rate, one thing too small for a feature but too good not to show, a few links, and one question for you.
A build graduates when it's functional, stable and safe enough to release, as is, to the world. Graduating means I release the code to everyone, or I hand the project to a team that wants to take it on. Who decides? It's me. It will always be me.
Once a build graduates: a monthly maintenance update and post-mortem, on a day that isn't Tuesday.
A paid tier comes later: early access to features, feature requests and test builds. The code and its security fixes are never behind it.
I'll also call it like it is: one of the reasons HubSpot is even on my mind, and others', is its recent layoffs. In early October it announced about 660 job cuts, roughly 7% of the company.
I'm not here trying to vilify their corporate decisions. I've been on both sides of the decision to let people go, and it's rarely an easy one.
But I'm familiar with the culture HubSpot professed while Katie Burke was Chief People Officer (she did a wonderful job in her tenure), the transition from co-founder CEO Brian Halligan to its current leadership, and the shift that followed her exit. It's a different company facing different challenges, and I'm not involved enough with their day-to-day decisions to know whether this reduction in force makes sense.
But I do know that, much like HubSpot, entrepreneurs, SMBs and enterprise customers also find themselves in a very different space, so they too should re-evaluate whether offerings like HubSpot's still make sense on the value proposition being offered.
Sales Hub Professional for a 15-person team runs about $17,700 in year one:
Tomorrow: what the imitation has to do, what's out of scope, how it gets judged, my prediction, and the first week.
Burn rate: about 36 hours (mine), 183.8 million tokens, $166.70 real money (the Claude upgrade, $100; this build's Supabase project, $10; cloud test computers, $39.25; and AI model tests, $17.45), 78 days left.
The ask
Reply with the software you'd most like to see duped next.
