The GTM teams getting the most from AI are building engines, not completing tasks
92% of Claude Code users say it saves time. That's the floor. The ceiling is what 67% report. Workflows that were previously impossible.
A survey of 200 GTM operators who use Claude Code or Cowork found that 92% say the product saved them time. That’s the expected result and the wrong thing to optimise for.
The more useful number: 67% said they built GTM workflows that were previously impossible. That’s a different category of outcome. Faster is incremental. Previously impossible is a structural change in what the team can do.
The split between the two groups comes down to what they’re building. The teams reporting time savings are using AI for tasks. Faster drafts, quicker research, summarised calls. Valuable, but bounded. The task completes, the time is saved, then the next task starts.
The teams reporting previously impossible outcomes are building GTM engines. When asked which use case had the biggest GTM impact, the top answer among Code users was GTM engines and prospecting. Signal-based outbound sequences, tools built for SDRs, messaging adapted dynamically by segment. These aren’t tasks that run once. They’re systems that run continuously.
The practical difference: a task-based approach to signal-based outbound means a rep gets an alert, reviews it, decides whether to act, drafts a message, and sends it. The AI might help with the draft. The sequence depends on the rep checking the alert.
An engine-based approach means the signal fires, the system enriches the account, scores the signal against the ICP, routes it to the right rep with a pre-drafted message and account context, and triggers a parallel ad sequence to the buying committee. The system runs whether the rep checks it or not.
The setup investment for an engine is higher than for a task. It requires real context engineering. Connecting live data sources, building workflow logic, defining routing rules. Most teams won’t do it. The ones that do get infrastructure that runs without manual input, improves as signals are correlated with outcomes, and compounds as more context is added over time.
The question worth asking when evaluating where to use AI: is the goal to do this task faster, or to make this motion run without someone manually driving it?