Status: Build 01, HubSpot Sales Hub Professional, is two weeks in.
Pre-alpha, ahead of where I expected, and 77 days left.
In July, Ian Phillips of Phillips Data Solutions wrote that a small business can replace HubSpot with a custom CRM, and that they'd done it in "roughly a week", on under $25 a month of infrastructure. His company sells custom CRMs, and the post is about HubSpot's Starter plan.
I think it's embellished, but I also think he knows it. It's a marketing piece, with hyperbole to catch your attention and the SEO of HubSpot's name. (Honestly, that's not so different from me putting HubSpot's name on this project instead of "Random CRM Dupe".) It's bold, overexaggerated and it serves his company's goals, but it's not completely rooted in lies. For the right users, there's truth to it. As with all things, the truth is in the details.
There's so much more to running a successful service than coffee-fueled vibe sprints and delusions of grandeur. It's still hard work, and it isn't sitting on a beach in Cebu mashing a "dangerously skip permissions" button. AI tooling has lowered the barrier to entry, but it still takes work, skill and knowledge. I'm old, so I've seen this before: HyperStudio, FrontPage, Dreamweaver, Joomla, Drupal, WordPress. We've claimed X was going to kill Y in this space since before Google existed.
So I'm going after Pro. Starter isn't that useful as a package. It's designed to give you just enough of a taste of what a CRM can do, with enough friction to nudge you up a tier. As a product manager, that makes complete sense to me, and HubSpot does it well. Pro is the first tier where I think HubSpot delivers real value over open source options like Twenty and Atomic (both great projects, worth a look if you're considering open source).
For 15 people that's about $17,700 in year one, and closer to $34,800 if they all send quotes, since HubSpot moved quotes into a separate Revenue Hub in June ($95 a seat a month on top, unless your account predates September 3, 2025).
This isn't for everyone on Pro. It's for the team that got forced onto it because they need X, Y and Z, but could live without A, B and C, to the tune of thousands of dollars. People willing to trade customer support, some features and an account manager for opex savings or flexibility. That's not to say Pro is a bad option. It's a different one.
To get there, it has to:
Beat Twenty and Atomic. They're mature and solid. If this doesn't beat them, why wouldn't you just use one and customize what you need? I'm comparing every feature now.
Work for more than Google. Right now it's Google Workspace only, and the business world is by no means dominated by Gmail. I can't call it a replacement until that's fixed.
Close the gap on Pro itself. Of HubSpot's 138 Pro features, 29 are built, 32 are partly built and 77 are missing, 25 of those on purpose (calling, payments, enrichment data, HubSpot's AI agents and app marketplace). The full table is at dupecrm.com/built, updated with every issue.
This is a brand new newsletter with no existing brand behind it, so I don't expect readership to skyrocket overnight, or thousands of teams to try this thing out. I'm not an influencer and I don't want to be one. I'm going to build cool things, run them, and type my nonsense thoughts into this newsletter. If people come along on that journey, neat. If they don't, that's also fine.
Success: one person hosts it themselves and uses it. Or 20 small businesses of 10 to 20 people pay for a managed version. Or it finds a home with other maintainers, or a company that takes it on under its own brand.
Fail: it's genuinely better than HubSpot, but unsustainable for a person or small team to run when buying HubSpot makes more sense. Saving $50 or $100 before you count your own time isn't a good deal, and given what HubSpot costs and the work involved, even $1,000 a year might not be. (Unless you already run internal tools. Then maintaining a solid open source tool might be no big deal, and that's a success.)
The code: a post-challenge thing. Handing it out early would count as a release, and if it isn't safe enough, it shouldn't graduate. The verdict comes January 19, with every number.
A one-man show rebuilding a massive SaaS offering is enough unreasonable expectation already. We're building business software under a goofy name like "Duped", not solving world hunger. It's not that serious.
If it has to be yes or no, it's a yes, with conditions.
Limited to Sales Pro, I think it's doable, mostly because I think HubSpot is using the wrong business model for it. Per-seat pricing is the wrong way to go for a sales CRM. If Nylas and Resend handle the annoying parts of email, this can be a win for AI against the seat pricing of that tier. (The Marketing product is where I think a solo dev would be outmatched. Well, that's an assumption. Maybe it's a future challenge.)
The catch: HubSpot can absorb seat-based economics as part of its moat, to raise stickiness and drive value to its marketing product. I don't have that option. Every seat I run adds a Nylas bill. And if HubSpot reprices this tier, and lowers its own costs to do it, it could crush me overnight.
The build is ahead of where I expected, but it by no means feels like I said "just build a HubSpot" and got the result I wanted. It's taken a fair amount of guidance from me, and my own experience with AI and non-AI development has helped a lot.
Day one, 16:20: the nine features I started with existed in code, with 163 automated tests, all passing.
Day one, an hour later: 50 minutes of clicking found 10 bugs those tests missed. Forms wiped what you'd typed when a save failed, times showed in the wrong time zone, and errors read "expected string, received null."
Day one, my inbox: about 2,450 of my emails were readable by "every teammate", because "share my email" had been built as "share my mailbox". Nobody else had an account, so it was only my email in front of my own eyes, and it was fixed that afternoon.
In my view, that's completely normal. That's what building a new product looks like, and it's why the first release of a feature isn't rolled out to users straight away, and why we use test accounts and take offensive and defensive security approaches to our own builds. It's pre-alpha: only I use it, it isn't a release candidate yet, and I wouldn't let a company run on it.
The hardest part is the lack of human QA testers. Development would be exponentially faster with people using it and sending feedback and bug reports. When there's a paid tier, release candidate access will be part of it, and I'll want to hear what you like, don't like, are worried about and find.
I put the demo up early because that's the experiment. I thought we'd get here around the end of month two. We didn't need that long, so it'd be a lie to tell you it isn't ready for eyes on it. It's safe enough to click on, break and have opinions about.
Yours alone: every visitor gets a private sandbox of made-up data.
Real: contacts, deals, pipelines, automations, reports, the forecast and undo, plus the AI features (30 requests per sandbox).
Simulated: the inbox, sending and booking. A public sandbox that sends real email would be a gift to spammers.
Not sure it fits your team? The "Is it ready for your team?" box on dupecrm.com/built asks how you sell today and tells you what's built, what's missing and what blocks you.
I'm not going to tell you what to judge it on. But if anything, go easy on speed. It runs on a small instance because I'm cheap. If it gets real traffic, I'll pay for a bigger one, and that'll show up in the real money. That's a good problem to have.
Burn rate: about 40 hours (mine), 186.5 million tokens, $176.70 real money (the Claude upgrade, $100; Supabase for this build and the demo, $10 each; cloud test computers, $39.25; and AI model tests, $17.45), 77 days left.
What it'd cost to run your own copy
A 15-person team at list prices: Supabase Pro $25, Vercel Pro $20, Nylas for 15 mailboxes $26.25, Resend $0, AI features about $1. About $72 a month, or $870 a year, against $17,700 for HubSpot. What that leaves out is your time, which is the whole question.
Too small for a feature
Automations you describe in plain words. One of my frustrations with our current CRM, Attio, is that its automations are still too hard for my non-technical users. I actually like Attio, but as a CPO, that complexity means the feature either goes underused, or I end up configuring things for people instead of them serving themselves.
Here you can just type what you want: "When a new deal over 10k is created, give the owner a task to book a kickoff call within 3 days." It writes the rule back in plain words, shows what it would have done over the last 30 days, and only then asks whether to turn it on. And it doesn't need an expensive model: three cheap ones got all 80 test requests right, at about two hundredths of a cent each.
Links
Ian Phillips, "Yes, an SMB can replace HubSpot with a custom CRM" (Phillips Data Solutions, July 3). The claim this build tests.
Jason Lemkin, "The 90/10 Rule for AI Agents" (SaaStr). They replaced a paid tool in about a day and a half, then found out what maintenance costs.
Retool's build vs. buy report (February 2026). 35% of the 817 Retool users surveyed have replaced at least one SaaS tool with something they built. Retool surveying its own users.
The ask
Try the demo, then reply: what's the one thing that would stop your team from switching?
Then keep an eye on it. I'll roll out updates periodically, and watching the built list will tell you whether this is worth your time. It's going to have ups and downs, successes and failures, but it'll let you gauge how it's going, and whether this has any chance of being useful by the end of month three.
