The Hidden Cost of AI Speed: Why 3 Teams Were Building the Same Thing
April 2026 · 5 min read
The bottom line
AI makes individuals faster — but it also makes duplication faster. At Groupon, I used AI to search across all our project management tools at once and discovered 3 teams independently building the same merchant notification system. The same tool that creates the problem is the best tool to detect it.
The problem nobody talks about
Everyone celebrates how AI makes individuals faster. Build a pipeline in an afternoon. Prototype an integration in one session. Ship a workflow without waiting for engineering.
But here's what nobody mentions: when anyone can build anything quickly, the probability that someone else is building the same thing goes up, not down.
The result: overlapping swim lanes, fragmented data quality, wasted engineering time, and solutions built by people who don't fully understand the domain — missing edge cases that existing systems already handle.
What I found at Groupon
I own the Merchant Lifecycle Engine at Groupon — the system that detects underperforming deals and triggers merchant outreach. To understand the full landscape, I ran 4 concurrent Claude Code agents across our tools simultaneously:
- → Agent 1: searching Asana across all projects for keywords like “merchant alert”, “notification”, “deal expiration”
- → Agent 2: querying BigQuery for actual alert volumes by type, channel, and region
- → Agent 3: searching GitHub for all notification templates in the codebase
- → Agent 4: fetching Jira tickets related to merchant alerting
5 minutes.That's how long it took. Here's what came back:
3 teams were independently building merchant notification systems with overlapping triggers — each solving a real problem, none knowing the others existed.
12 tasks across 4 Asana projects owned by 4 different teams were doing versions of the same work.
The BI team had already built a Keboola data connection to our alerting service — but nobody was using it. It was sitting there, functional, invisible.
The Travel team was building their own deal expiration alerting, duplicating logic that already existed in the main system.
This wasn't malice or incompetence. Each team was solving a legitimate problem for their stakeholders. They just didn't know the others were doing the same thing.
Before: siloed
4 teams, 4 systems, overlapping triggers, no shared data model
After: unified
Unified Alert Catalogue
→ One trigger logic, multiple channels (SMS, email, Chat, tasks)
→ One data pipeline, shared by all teams
→ One catalogue documenting all 40+ alert types
→ One owner, with engineering support
Consolidated in the Unified Alert Catalogue
Why this happens
AI lowers the barrier to build.That's the whole point — anyone can prototype a solution in an afternoon. But the flip side is that more people are building independently, without visibility into what already exists.
Project management tools are siloed.Team A uses Asana. Team B uses Jira. Team C uses a spreadsheet. Even when everyone's on the same tool, projects are separated — nobody searches across all of them.
Institutional knowledge lives in people's heads. The person who built the original solution left, or moved to a different team, or is too busy to join your standup. The documentation — if it exists — is in a wiki nobody reads.
The builders often aren't the domain experts. A well-intentioned engineer or analyst builds a pipeline, a dashboard, or an alerting system — but without deep domain knowledge, they miss edge cases, use different naming conventions, hardcode values instead of connecting to real configs, and create data quality problems that compound silently.
The fix: AI-powered cross-team mining
Take the tool you already use for alignment — Asana, Jira, Notion, whatever — and use AI to search across ALL of it at once.
# Search across all Asana projects:
"Search all projects for tasks containing 'merchant alert',
'notification', 'deal expiration', 'merchant outreach'"
→ Found 12 tasks across 4 projects owned by 4 teams
→ Consolidated into one unified catalogue
This is what MCP servers enable. You connect Claude Code to the Asana MCP, the Jira MCP, or both — and run keyword searches across everything. Not just your project. Not just your team. Everything.
The search takes seconds. The consolidation conversation takes longer — but at least you're having it before three teams ship three different systems.
What you find when you look
- Duplicate workstreams — multiple teams building notification systems, data pipelines, or dashboards for the same underlying data.
- Abandoned infrastructure — connections, pipelines, and integrations that someone built, that work, but that nobody uses because nobody knows they exist.
- Conflicting implementations — team A sends SMS, team B sends email, team C posts to Google Chat, all triggered by the same event but with different logic, different data sources, and different edge case handling.
- Data quality erosion — non-domain-experts building data pipelines that use different naming conventions, miss validation rules, or hardcode values that should come from config services. This compounds silently until someone tries to build a dashboard across all of it and nothing matches.
How to implement this
- Connect your project management tool to AI. Asana MCP, Jira MCP, or direct API. The connection takes minutes — the hard part is deciding to do it.
- Build a keyword search that runs regularly.Weekly is enough. Search for the terms that define your domain — “merchant”, “alert”, “notification”, “pipeline”, whatever is relevant to you.
- Flag overlaps to a central person. An ops lead, chief of staff, or technical program manager who can make the call: consolidate or let it diverge intentionally.
- Before any new project starts, search first.“Has anyone else done this?” should be a 30-second check, not a week of meetings.
- Maintain a living catalogue. We call ours the Unified Alert Catalogue — a single document listing every active workstream, its owner, its status, and how it connects to the others. Updated when new workstreams are discovered.
AI doesn't just make you faster. It makes duplication faster too. When the barrier to build drops, more people build — and the chance of invisible overlap goes up with every new project.
The same tool that creates the problem — AI — is the best tool to detect it. You just have to point it at your own organization, not just your customers.