The Failover Economy
Here is how I attach a different AI brain to my body of work, and it turns out I’m not alone.
One desk’s story, checked against 30 months of corporate card data.
The dread
My primary AI is Claude Code. When the usage meter creeps toward 100%, my workday used to just… end.
Then I read Adam Schoenfeld describing the same moment as a shrug: he flips to a second agent and keeps working.
This is what I did about it, and the data that says I’m not alone.
June 2024
On Ramp’s corporate cards, OpenAI appears on 30.7% of US businesses.
Anthropic is at 5.7%.
One incumbent, one asterisk. A market that looks decided.
May 2026
Two years later the lines cross: Anthropic 41.0%, OpenAI 39.5%.
The crossover made headlines. It is also not the story.
The story
Add the two shares: 81.9%. Total AI adoption: 54.9%.
The arithmetic forces a floor: at least 27% of all businesses pay both. Two years ago that floor was 3%.
Businesses stopped picking a winner. They built failover.
What I did about it
As of today, this desk is in the violet band. The whole move: one rulebook symlinked so any agent reads it (ln -s CLAUDE.md AGENTS.md), shared memory files, a map of which tools work from any agent, a handoff file for mid-task flips, install Codex, then the acceptance test: read my rules and recite them. It did.
Ten minutes of setup. The dread is gone.
The full walkthrough: Make Your AI Agent Replaceable.
AI-adopting US businesses now provably pay both OpenAI and Anthropic. The chart counts businesses; as of today, one more desk runs the same way.
Method, honestly
Ramp publishes the share of its ~50k US business customers with paid subscriptions to each AI vendor, and the share paying for any AI. Vendor shares overlap. The violet floor is inclusion-exclusion: OpenAI% + Anthropic% can only exceed total adoption% if the difference pays both, so the true both-vendors number is at least that difference, and likely higher. Ramp’s base skews toward card-forward companies. The data below is embedded in this page as JSON: view source, take it, check the math.
Companion essay: Make Your AI Agent Replaceable, the ten-minute setup that puts this chart into practice at desk level.