You're Paying for the Same AI Five Times

Run this exercise: pick one engineer on your team and list every AI assistant your company currently pays for on their behalf. A typical answer in 2026 looks like this — an assistant in GitHub, an assistant in the IDE, an assistant in Jira and Confluence, an assistant in Slack, an assistant bundled into the cloud platform, and a standalone chat subscription somebody expensed. Now ask what each one is for, and watch the answers converge: they summarize things, they draft things, they answer questions about the codebase, they write code.

You are paying five vendors for three capabilities. So is almost everyone.

How we got here

Nobody chose this. It accumulated. Every SaaS vendor you already pay spent the last two years bolting an assistant onto their product and folding it into the renewal — sometimes as an upsell, sometimes "free" with a price increase that wasn't framed as one. Meanwhile, teams ran their own procurement: engineering bought a coding agent, support bought a ticket assistant, someone in marketing bought a chat subscription with a corporate card. Analysts tracking AI spend describe exactly this pattern: overlapping subscriptions, multiple concurrent pilots for the same use case, and duplicate spend that nobody can see because it's scattered across departmental budgets and bundled line items.

The market is now correcting. Roughly two-thirds of technology leaders say they're actively cutting AI vendor portfolios this year, with most targeting about a 20% reduction. The consolidation wave is coming to your budget whether you steer it or not. My argument: steer it, because consolidating by inertia — keeping whatever's bundled with your biggest contract — routinely keeps the wrong tools.

Bundled is not the same as best

The trap in an overlap audit is resolving every duplicate toward the bundle, because the bundled assistant looks free. It isn't free — you're paying for it in the renewal — but more importantly, "included" says nothing about "good at your job." The delta between the best and the median tool on a given workflow is enormous right now, and it's largest exactly where the money is: agentic coding work. A bundled assistant that autocompletes politely and a dedicated coding agent that completes a migration overnight are not interchangeable line items, even though they land in the same "AI assistant" category in a spend report.

The pricing models are also diverging in ways that make naive comparison misleading. Some tools are seat-priced; agentic tools are increasingly consumption-priced — Devin bills in compute units, Claude Code draws down usage that scales with how hard you push it. Comparing a $19 seat to a consumption bill isn't apples to apples; you have to compare cost against the work actually delivered. A tool that costs 4x per month and replaces 30 hours of engineering time is the cheap one.

The audit, concretely

Here's the shape of the exercise I'd run, and it fits in two weeks.

First, inventory by capability, not by vendor. Build a simple grid: summarization, drafting, code generation, agentic/background coding, codebase Q&A, meeting notes, search-over-internal-docs. For each capability, list every tool you pay for that claims it, what it costs (unbundle the renewal math where you can), and actual usage — most of these tools have admin dashboards now, and usage data kills more zombie subscriptions than any argument.

Second, name a primary tool per job. Not per team — per job. One primary for agentic coding, one for in-IDE assistance if you keep it separate, one for knowledge search. Secondaries need a written justification: a real capability gap, not a preference. This is also where you check the interoperability escape hatch — tools that speak MCP can share context and integrations, which lowers the penalty of consolidating and the cost of switching later.

Third, kill with a calendar, not a memo. Every tool that loses its category gets an end date tied to its renewal. The savings are real, but the bigger win is focus: enablement effort, shared configuration, and internal patterns concentrate on the tools you actually kept, which is what moves adoption past the enthusiast crowd.

What not to consolidate

A caution from the other direction: don't consolidate your way out of leverage. Keeping one secondary coding agent in genuine use — a real workload, not a shelf license — does two things. It keeps renewal negotiations honest, and it keeps you calibrated on a market that still reshuffles every six months. The goal isn't the minimum number of vendors; it's zero unowned overlap. Every tool has a job, every job has a primary, and somebody can explain the exceptions.

The takeaway

The AI line item is now big enough that CFOs are looking at it, and the first pass at cutting it will be made by whoever shows up with a spreadsheet. Better that it's you, with a capability grid and usage data, than procurement with an alphabetized vendor list. Audit the overlap, pick primaries deliberately, keep one hedge, and put the savings into enablement for the tools that won. Paying once for something great beats paying five times for the same mediocre summary.