We’re now nine months into running SaaStr AI as three humans, one dog, and 20+ AI agents. We’ve vibe-coded 8+ apps into production. We’ve bought dozens of third-party AI tools. And in the last two weeks alone, we built two new apps — one internal, one external-facing — that taught us some hard lessons about when to build, when to buy, and why the old rules are changing fast.
Here’s the updated playbook.
The 90/10 Rule Still Holds — But With a New Exception
Our rule from Day 1 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 our marketing automation platform. We’re not recreating our outbound SDR agents. All of those have solid third-party tools that do the job well. And when you’re a team of three humans managing 20+ agents and 8+ apps, you simply don’t have the bandwidth to take on compliance, security, data warehousing, and everything else that comes with building from scratch.
But here’s the new wrinkle: we now also build when a tool we’re paying for has zero AI functionality.
That’s the line. If it’s February 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.
What We Built: Two Apps, Two Very Different Reasons
1. AI VP of Marketing (Internal Tool — “10K”)
We couldn’t find a marketing orchestration tool that could do what we needed. The problem wasn’t just features — it was data. We had 10 years of proprietary data on what’s worked for SaaStr across sponsors, tickets, customers, and media. No new hire could process all of that. No off-the-shelf tool could ingest it.
So we built our own.
The AI VP of Marketing ingests all our historical data, connects to Salesforce via APIs and Zapier webhooks, and outputs a six-month marketing plan with daily actionable to-dos. Every morning, it tells us where we stand against our attendance and revenue goals, what campaigns to run, what copy to use, how much to spend on ads, and where to adjust.
It even roasts us when we’re behind.
This was a classic “10% build” — internal-facing, low risk to external users, and no off-the-shelf alternative existed. We gave the AI a day to ingest all our old data, analyze what’s working, run competitive analysis, and come up with a plan. It delivered.
2. SaaStr Sponsors Portal (External-Facing — Replaced a Paid Tool)
This one was different. We were paying for a sponsor portal tool. It did some of what we needed. 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 AI 101. And it wasn’t there.
The old tool also had broken persistence on single sign-on — if Amelia submitted something and I logged in, I couldn’t see what she did. And every time a sponsor needed to add someone to their account, a human on our team had to manually invite them. In 2026.
So Amelia gave herself a constraint: one day to get single sign-on working. If it works, build the rest. If not, go back to the old tool.
She got it done. Here’s how.
How Claude Co-work Changed the Build vs. Buy Calculus
The SaaStr Sponsors portal probably doesn’t get built without Claude Co-work. Here’s why:
Step 1: Write the spec with Claude. Amelia gave Claude the URL of the existing tool’s front end and said: “Look at this, write me a spec for a replacement, and make sure it has AI features baked in.” She also specified that single sign-on was non-negotiable and that the app needed to stay mostly open (since sponsors often work with agencies who need access too), gating only actions like task submissions and booth selection.
This is a critical step. Write the spec first. It bounds the project, prevents the vibe-coding agent from going rogue on features you don’t need, and gives you a clear contract for what “done” looks like.
Step 2: Build in Replit / Lovable with the spec. She took the Claude-generated spec into Replit and started with the hardest part — true single sign-on using Clerk. Not fake SSO. Real, persistent, org-wide SSO where anyone from a sponsor’s domain could log in and see their company’s data.
Step 3: Use Co-work as the glue. Because Claude Co-work sits on your machine and in your browser, it could see everything — the old tool, the contracts folder, the registration platform. So Amelia told it: “Go through every sponsor contract in this folder. Extract the company, URL, number of sponsor passes, speaking slots, and signing date. Format it so I can feed it to the Replit agent.”
An hour later, Claude had processed all 150+ contracts.
Then: “Now go into Bizzabo (our registration platform) and create unique sponsor codes for each company.” Twenty minutes later, done. Promo codes, nomination passes, everything.
That entire workflow used to take a human a full day. Claude did it in a couple hours.
Step 4: Test obsessively. This was Amelia’s first time implementing SSO. Ever. She’s a go-to-market person, not an engineer. So she tested everything — signing in with different emails, submitting tasks, checking persistence across accounts. She sent it to 12 sponsors first. Watched the submissions come through on the super admin backend she’d also built into the app. No issues.
Step 5: Ship it. SaaStrSponsors.com is live. External-facing. Managing millions in sponsor revenue. Built by a non-engineer in roughly a day and a half.
What Anton’s Data Tells Us About Where to Start
Anton (from Replit) posted data on what people are actually vibe-coding. The ranking is telling:
- Rapid prototyping — Quick mockups, landing pages, case study sites. Low risk, high value, great entry point.
- Internal tools — Like our AI VP of Marketing. Internal-facing, proprietary data, no off-the-shelf fit.
- Automating one-off deliverables — Taking a call recording, piping it through Zapier into Replit or Lovable, and spitting out a custom website for a prospect instead of a generic deck. We do this constantly now. Low risk, high personalization.
- Replacing simple SaaS apps — This is where the sponsor portal sits. You don’t get here on Day 1. This was after nine months of building multiple production apps and websites.
If you’re just starting out, don’t try to rebuild a customer-facing portal with SSO. Start with prototypes and internal tools. Build your muscle. Then go after the replacements.
The Catch: Maintenance Is Real and It’s Compounding
Here’s what nobody talks about on social media: every app you build is an app you now have to maintain.
I spend a significant chunk of my day maintaining vibe-coded apps. Amelia does too now. And the temptation to keep fussing with a working app — “what if we added this feature, what about that widget” — is real and dangerous, especially when your time is worth $1,000-$2,000 an hour (Claude’s estimate of Amelia’s hourly value given our lean team structure).
There’s no economic justification for building these apps based on software cost savings. You can buy a sponsor portal tool for $5K-$10K. The justification is purely that the off-the-shelf options are so far behind that they’re actively hurting your business.
The next frontier for vibe coding isn’t building — it’s maintenance. Who maintains 10 apps and 20 agents for the next five years? Right now, the answer is us. That’s not sustainable. We’re at our max.
Security: Not Solved, But Getting Better
For public-facing vibe-coded apps, here’s our approach:
The platforms (Replit, Lovable, etc.) now do basic security scans at launch — checking for exposed private keys, common vulnerabilities. That helps.
Beyond that: ask your agent to do the deepest security audit possible on your app at least once every month or two. They’ll find a lot. It’ll take hours to fix everything, and the fixes will probably break other things. Each security update cycle takes us about a week to fully recover from. But it’s better than most low-end outsourced dev shops would give you.
It’s not a fully solved problem. But it’s manageable for simple apps that aren’t storing highly sensitive data.
The Jaw-Drop Test: The Only Question That Matters Now
Here’s the harder truth for B2B founders.
Shopify just reported growing 30% at $12 billion in revenue. The stock dropped 10.4% that day. Atlassian is re-accelerating at $5 billion in revenue and trades at 4x ARR. The world we grew up in over the last decade doesn’t work at those multiples.
You can hope the old days come back. Most startups I get investor updates from that aren’t AI-forward are basically doing that — hoping. But hope isn’t a strategy.
The real question is: does your product still make your customers’ jaws drop?
At some point, your product was a jaw-dropper. That’s how you got your first customers. They didn’t buy you because you were 1% better. They bought you because you fundamentally changed how they worked.
Is that still true? In the age of AI, is your product as jaw-dropping as Claude? If not, that’s why you’re decaying. And you need to find that jaw-dropping AI moment — the one where you can’t believe what was possible six months ago.
We added a new AI SDR tool this week — Monaco, from Sam Blond (former CRO of Brex). It finds your ICP, calls them, and sets up confirmed meetings directly in your calendar. You don’t run cadences. You don’t do sequences. It just books real meetings with real humans. That’s a jaw-dropper. That’s worth paying $50K-$100K for. A crappy project management tool with no AI for $200/month? That’s a cancelled subscription we already ghosted.
There is no middle ground anymore. You’re either at the cutting edge of AI, or you’re getting displaced. The only question is speed.
The Bottom Line
If you haven’t started vibe coding yet: Now is the best time. The tools are radically better than even two weeks ago with the new models and Claude Co-work. Next week will be even better.
If you’re evaluating build vs. buy: Stick to the 90/10 rule. Buy everything you can off the shelf. But if a tool you’re paying for has zero AI features in 2026, that’s your signal to start building.
If you’re a B2B founder: Use your own product today. Is your jaw dropping? Would you renew? Would you pay more than last year? If the answer is no, you know what you need to do.
If you’re managing agents: There is no set-and-forget. Every morning starts with an hour catching up with your agents. Every week they get a little easier. But mediocre inputs get mediocre outputs. Put in the time or don’t bother.
We’ll be teaching all of this hands-on at SaaStr Annual 2026, May 12-14, with a full SaaStr Deploy day on the 12th dedicated entirely to deploying agents and building your own B2B app. Come build with us.
You can see all the agents and tools we use at saastr.ai/agents.
