Dear SaaStr: I keep hearing about “vibe coding” and founders building their own internal tools with AI. But we already pay for a bunch of B2B products. When does it make sense to vibe code something yourself vs. just buying an off-the-shelf product?
Great question, and one we have strong opinions on because we’ve lived it. We run SaaStr with three humans, one dog, and 25+ AI agents. We’ve vibe coded 10+ production apps. We’ve also bought dozens of third-party tools. And we’ve made mistakes in both directions.
Here’s what we’ve learned.
Start With the 90/10 Rule
Our rule from day one has been simple: buy 90% of what you need off the shelf, and only build the 10% where no solution exists.
We’re not trying to vibe code our own CRM. We’re not rebuilding Salesforce or HubSpot. We’re not recreating our outbound SDR agents. All of those have strong products with real engineering teams behind them. When you’re a team of three humans managing 20+ agents, you don’t have the bandwidth to take on compliance, security, data warehousing, and everything else that comes with building core infrastructure from scratch.
Most founders get this wrong in both directions. Some try to vibe code everything. Some refuse to build anything. The 90/10 split is about discipline.
Buy the platform. Only build what you really ,truly need. And can’t find.
But We Added a New Exception in 2026
The 90/10 rule still holds. But we added a wrinkle: we now also build when a tool we’re paying for has zero AI functionality.
That’s the new line. If it’s 2026 and your product doesn’t have a single AI feature, not even a lightweight one, that’s when we start looking at replacing you. Not because we want to. Because we have to.
Let me give you two real examples. One where we built because nothing existed. And one where we built because what existed wasn’t good enough anymore.
Example 1: AI VP of Marketing (We Built Because Nothing Existed)
We looked at every AI marketing agent on the market. Demoed a bunch. Really tried to buy our way out of this one.
But here’s the thing: the AI marketing agent tools out there all mainly just write mediocre content for you. Blog posts. Social captions. Email copy. We already have more content than we know what to do with. SaaStr has 5,000+ pieces of content. 20 million words. Content generation isn’t our bottleneck.
The real problem is orchestration. Strategy. Knowing what to do, when to do it, and executing against a coherent plan every single day across every channel. Not a single AI marketing tool we found could do that.
We also knew from experience that every time we tried to onboard a human with all our data, they got overwhelmed. Ten years of proprietary data on what’s worked across sponsors, tickets, customers, and media. No new hire could process all of that. No off-the-shelf tool could ingest it.
So we reluctantly built our own. We nicknamed it “10K” because it has two jobs: get us to 10,000 attendees for SaaStr Annual, and help drive $10,000,000 in revenue for the year.
What 10K does isn’t that complicated per se, but it’s comprehensive in a way no human on our team ever was. It uses Claude to do deep analysis of all our data for the past 4+ years. Every campaign. Every conversion rate. Every email open rate. Every sponsor interaction. Every registration pattern. That analysis feeds into an app we built on Replit that has designed every campaign, every offer, for each single day of the year. Not quarterly themes. Not monthly initiatives. Daily executable tasks.
When you click into any given week, it tells you the emails to send, what to do with your AI SDR, what to post on social, how much to spend on LinkedIn ads and what the ads should say, which sponsors to reach out to, and where you’re falling behind. It updates in real time. New data comes in, the plan adjusts. A campaign underperforms, it recalibrates. It’s not a static spreadsheet someone made in January and forgot about by March.
I talk to 10K every day. “Where are we? What should we be doing today? Where are we falling behind?” Sometimes I push back. It once suggested a campaign I didn’t think was urgent enough. We debated it. It looked at the data, looked at my points, and agreed to change course.
It’s not always right. But it does more than any human on our team ever did in this role. Not because our team wasn’t talented. But because no single human can hold 4+ years of campaign data in their head while simultaneously optimizing across 15 channels in real time.
This was a classic “10% build.” Internal-facing. Low risk to external users. And no off-the-shelf alternative existed. The gap between what the market offered (content generation) and what we needed (full marketing orchestration grounded in proprietary data) was the entire value.
Example 2: Sponsor Portal for Customer Onboarding (We Built Because What We Had Fell Behind)
This one was different. We were paying $10,000+ a year for a niche sponsor portal tool. It did the basics: let sponsors log in, manage booth details, upload logos. Fine.
But it had zero AI. None. Not even basic enrichment like “put in a company name and auto-fill their address, employee count, and description.” That’s table stakes in 2026. And it wasn’t there.
Every time a sponsor needed to add someone to their account, a human on our team had to manually invite them. We needed sponsors to pick their booth from an interactive map. We needed better flows for uploading assets and managing complimentary passes. We needed it to work the way our event actually works, not the way some generic template assumed events work.
And every time we asked for features? Either on a roadmap that never materialized, or a custom development fee on top of the subscription.
This is the fundamental challenge with niche vertical software: the TAM is small, so the engineering team is small, so the product evolves slowly. It’s not the vendor’s fault. The economics just don’t support a world-class engineering team for a few hundred customers.
So Amelia gave herself a constraint: one day to get true single sign-on working. If it works, build the rest. If not, go back to the old tool.
She got it done. And she’s a go-to-market person, not an engineer. She used Claude to write the spec first, giving it the URL of the existing tool and saying “write me a spec for a replacement with AI features baked in.” Then she built it in Replit, starting with the hardest part: real, persistent, org-wide SSO using Clerk. Not fake SSO. Real SSO where anyone from a sponsor’s domain could log in and see their company’s data.
Then came the part that would have taken a human a full day: processing 150+ sponsor contracts to extract company names, URLs, sponsor passes, speaking slots, and signing dates. Claude Co-work did it in about an hour. Then it went into Bizzabo and created unique sponsor registration codes for each company. Twenty minutes.
It’s now live. External-facing. Managing millions in sponsor revenue. Built by a non-engineer in roughly a day and a half. The old vendor lost us forever. They don’t even know yet.
But notice: this was not a Day 1 project. We were nine months into building production apps before we attempted replacing an external-facing paid tool with SSO. Build the muscle on internal tools first. Then go after the replacements.
The Decision Framework
After building 10+ apps and buying dozens of tools, here’s how I’d break it down.
Vibe code it when:
- The tool you need simply doesn’t exist as a product. That was our AI VP of Marketing. No vendor had built marketing orchestration grounded in proprietary data with daily executable tasks. The gap between “available” and “needed” was so wide that buying wasn’t an option.
- Or when a tool you’re paying for has fallen so far behind that it’s actively hurting your business. That was our sponsor portal. Zero AI in 2026. Broken SSO. Manual processes that should have been automated years ago. When your vendor stops innovating, you stop renewing.
- Or when speed matters more than perfection. I can go from idea to working prototype in 90 minutes. No vendor evaluation. No sales calls. No procurement process. No 30-day trial. No implementation timeline. If you’re testing a hypothesis or solving a problem that might not even be a problem in three months, vibe coding gets you to an answer fast.
Buy it when:
- Multiple well-funded companies compete in the category with real AI-focused engineering teams. You are not going to out-build HubSpot’s CRM in a weekend. You’re going to build a worse version and spend six months maintaining it instead of running your business. The test is simple: is this product someone else’s full-time job? If yes, pay them.
- Or when reliability and compliance are critical. If the tool touches customer data, handles payments, requires SOC 2, or needs to work at scale without you babysitting it, buy it. Vibe coded apps are great for internal tools and community-facing utilities. They are not great for anything where downtime costs you real money.
- Or when integrations with the rest of your stack are critical. Our CRM connects to everything. Our marketing automation connects to everything. Rebuilding those integration layers would take longer than building the app itself. Although in fairness, Replit and Lovable have more and more integrations off the shelf in them.
The Biggest Catch: Maintenance Is Real. And a Big Deal.
Here’s the important part that is under-discussed. Every app you build is an app you now have to maintain. I spend a meaningful chunk of my week maintaining vibe coded apps. Amelia does too. And the temptation to keep fussing with a working app, “what if we added this feature, what about that widget,” is real and dangerous when your time is worth far more per hour than the software costs per year.
There’s no economic justification for building based on software cost savings alone. You can buy these tools for $5K-$10K a year. You only build when the off-the-shelf options are so far behind they’re hurting your business. Or when nothing on the shelf does what you need at all.
We’re at our max right now. We can’t take on more apps to maintain. That’s a real constraint that most vibe coding evangelists on Twitter don’t mention.
Where to Start If You Haven’t Built Anything Yet
Based on what we’ve seen and what Replit’s data shows about what people actually build:
- Start with rapid prototypes. Landing pages, case study sites, quick mockups. Low risk, high value, great entry point.
- Then move to internal tools. Like our AI VP of Marketing. Internal-facing, proprietary data, no off-the-shelf fit. If it breaks, only your team is affected.
- Then try automating one-off deliverables. We take customer data, pipe them through Gamma and into Replit, and spit out both customer decks and a custom website for a prospect instead of a generic deck. Low risk, high personalization.
- Only then should you try replacing a paid tool. That’s where the sponsor portal sits. Don’t get here on day one. Build the muscle first.
The Real Answer
The question isn’t “vibe code or buy.” The question is: where has the market failed you?
Every company has 10-20 little workflows that are too specific for any vendor to solve and too small for any engineering team to prioritize. That’s the 10%. That’s where you open Replit at 9pm, describe what you want, and have something running by 10:30.
And in 2026, there’s a new category in that 10%: tools you’re paying for that have zero AI features. If your vendor hasn’t shipped AI by now, they’re probably not going to. And the gap between what you need and what they offer is only going to widen.
Buy the 90%. Vibe code the 10% where the market has failed you. And if you’re a B2B founder reading this, ask yourself: would your customers vibe code a replacement for your product? If the answer is yes, you know what you need to do.
