Filterbuy builds custom apps in Claude Code instead of buying off-the-shelf SaaS

"We build those and then we iterate, and that allows us to not just move fast and reduce costs—which was never the goal—but really build software that works for us as we want. So it’s custom software." — Tiago Almeida, VP of Technology, Filterbuy

Challenge

Filterbuy is a $260 million domestic manufacturer. It makes air filters and ships them across the country, a practical business built because people need the product. Founder and CEO David Heacock wants AI to make Filterbuy's people more effective. Using it to cut costs misses the point. He treats it as infrastructure for the work: tighter schedules, faster decisions, and fewer things missed.

The software they had was off-the-shelf. It was one size, built for the masses, and the teams worked around it. Each job needed a shape the product did not have. Building an app that fit used to take at least 10 people and a full year. Filterbuy was not going to staff that for every workflow.

"I hate when people say, I think they will want this. Don't. No. Don't do it. When they need it, you just build on the fly." — Tiago Almeida, VP of Technology, Filterbuy

Solution

Filterbuy uses Claude Code to make the software the company runs on, instead of buying it. David Heacock's long-term goal is what he calls a business operation system: everything Filterbuy-related, built, controlled, and customized inside the company. A request to buy software now goes to engineering before anyone signs. They evaluate whether to build it. Filterbuy partnered with Remotely.Works to staff the engineers who own these apps. The team builds in Claude Code on Claude Opus, with Claude review agents in the loop before an engineer signs off. The suite runs to 10 internal apps, each scoped to one job, including the finance app, the app for the B2B sales team, the customer support app, and the analytics app.

Some of those apps take the place of a product Filterbuy was paying for. A large product has to work for many companies, so it stays generic. Building lets Filterbuy customize the app to the job. The CRM that replaces Pipedrive is ready to release, and Pipedrive stays until that cutover. One engineer built the finance app that syncs Stripe to QuickBooks instead of a third-party invoicing tool. Customer support has replaced Five9 for live chat. Voice is still to cut over, and then Five9 comes off. Others in the suite are new. Analytics has an MCP so a marketer can ask Claude for the data instead of exporting it. Most apps also get an AI chatbot. On customer support, Claude reads the inbound email, pulls the customer context, and drafts the reply.

Each repo is moving to the same dev workflow. An engineer investigates, writes the spec, then runs that repo's hooks so the skills are the same no matter who opens the project. One engineer owns the app. A second joins for the non-critical work when a deadline is tight. More than two starts costing ownership, so they stop there. Claude Code generates the code. Review is the long step. The team will spend two or three hours on a review, with Claude review agents in the loop, because that is what keeps the quality high and maintenance low. The engineer reviews before it ships.

"I call everyone like product engineer nowadays. It's not much more software engineer. And why I'm saying that is because code is more or less solved with AI. So what really matters to me and to the team is understanding the context of what we're building, understanding the business, understanding what we're solving, what is the problem." — Tiago Almeida, VP of Technology, Filterbuy

How it works

1. A department brings the problem. Engineering defines the solution. A pod lead, who is an engineer, owns the relationship with that department.
2. An engineer owns the app and builds it in Claude Code. A second engineer joins only when the deadline needs it.
3. They build only what solves the problem. A feature stays unbuilt until someone has a clear unmet need for it. Tiago's example is the analytics MCP: one marketing director could not get what she needed inside Claude, so they built it for her, then expanded access.
4. The engineer reviews the generated code before it ships. Review is allowed to take longer than the writing, keeping quality high and maintenance low.
5. If the app replaces a paid tool, they turn the old one off. They stop paying for a product that does not fit, and avoid maintaining a second system on top of it.

Impact

• Builds apps 6-12X faster: What once took at least 10 people and a year now takes 1–2 engineers and 1–2 months. That is the average time to build an app in the suite. The B2B sales app was the long one: two people, three months. The newest apps reached production in 8 to 11 days.
• Merged pull requests went from 138 a month in June 2024 to 1,104 in August 2026.
• The teams do their jobs in these apps day in and day out. 40 sales reps use the app built for the B2B sales team. 20 customer support employees use the customer support app every day to do their core job.
• Each app is tailored to one core job, so it does that job better than a product built for many companies. Customer support has replaced Five9 for live chat. One engineer replaced the invoicing tool with the finance app that syncs Stripe to QuickBooks. Pipedrive stays until the internal CRM, which is ready to release, takes over.