The marketing ops case for building your own tools in Replit Agent

I spent the weekend rebuilding three tools my team had been paying for. Total SaaS cost eliminated: about $340 a month. Time to build: roughly six hours across the three. None of them are pretty. All of them work.

The combo doing the work is Replit Agent V2 with Claude 3.7 Sonnet underneath. I want to talk about why this specific pairing matters for marketing operators, and where I think most people will mess it up.

The category that just collapsed

There’s a class of marketing software that exists because nobody on the team can code: ROI calculators, UTM builders, simple dashboards that pull from one API, lead scrubbers, CSV transformers, embeddable quizzes, internal SEO scripts that hit Search Console and spit out a report. Every agency has paid $49 to $199 a month for some version of these for years.

That category is done. Not “disrupted.” Done. A marketing operator who can describe what they want in plain English can ship a working version in an afternoon. The agent handles the file structure, the deployment, the database, the auth if you need it. You describe outcomes. It writes code.

What changed isn’t that AI can write code. It could do that two years ago. What changed is that Claude 3.7 is good enough at reasoning through multi-file projects that the agent loop actually converges instead of spiraling. Replit’s team made a bet on Anthropic models back in mid-2024 and it’s paid off because the coding quality gap between Sonnet and everything else stayed wide.

What I actually built

Three things, in order of usefulness:

The first was a lead enrichment dashboard. Paste in a CSV of company names, it hits a couple of APIs, returns a sheet with industry, headcount estimate, and a fit score based on rules I defined. Replaces a Clay workflow I was paying for but barely using.

The second was an internal SEO audit tool. Give it a domain, it crawls the sitemap, checks title tags and meta descriptions against length rules and keyword presence, flags duplicates. Nothing fancy. But it runs on my schedule and stores history, which the $99/mo tool I was using made annoying.

The third was a UTM builder with a team-specific schema baked in. Just a form that enforces our naming conventions. Took 40 minutes. Saved me from explaining utm_campaign rules to a new hire for the hundredth time.

None of these are products. They’re internal tools. That distinction matters.

Where the wheels come off

The agent is not magic. A few honest notes from breaking things repeatedly.

It struggles when you change your mind mid-build. If you tell it to build a Postgres-backed app and then halfway through say “actually let’s use SQLite,” it will try, and the result is often a mess of leftover references that you’ll spend an hour cleaning. Start with the data layer decided.

It also gets cocky on integrations. Connecting to Google Search Console or the HubSpot API works, but the OAuth flow setup is where I lost the most time. The agent will happily write the code and then leave you to handle credentials in ways that aren’t obvious. Budget time for the auth dance.

And it cannot read your mind about scale. A tool you build for yourself runs fine. The same code shared with ten teammates may fall over because nobody wrote it expecting concurrency. Fine for internal use. Do not sell what you build here.

The skill that actually matters now

The bottleneck has shifted. It used to be: can you code? Now it’s: can you specify clearly what you want?

A marketing operator who is good at writing briefs is, it turns out, good at prompting a coding agent. The skill of decomposing a fuzzy goal into concrete requirements, defining inputs and outputs, knowing what edge cases matter, that’s the new moat. People who’ve run media buys or managed agencies have been doing this work for years. They just didn’t know it was transferable.

The people who will get the most from this are not the ones who already code a little. It’s the ones who’ve been writing tight project specs for vendors. They already think in the right shape.

What this changes for budgets

If you run a marketing team, do an audit this quarter. List every SaaS tool under $200/mo. For each, ask: could one person on my team rebuild the 20% of features we actually use in a day with Replit? For roughly a third of them, the answer is yes. That money goes to either better tools at the high end or to your own salary line.

If you’re a solo operator or running a small agency, the math is even cleaner. A Replit subscription plus Claude API access runs you a fixed monthly cost that replaces a dozen subscriptions if you’re willing to build. The catch most readers will miss: you have to maintain what you build. The hidden value of SaaS is that someone else patches it when an API changes. Build only the tools whose maintenance cost you understand and accept. Everything else, keep buying. The discipline isn’t in building more. It’s in knowing which things are worth owning.