Anthropic on Wednesday released a set of reports that show managers what a team did with Claude Code, what it cost, and where the work stalled, TechCrunch reported on 28 October 2026. A manager picks a date range and sees the output, the spend, and the points where workflows broke down, all built from internal analytics and employee surveys. The reports also suggest ways to get better results. By default, Anthropic says, the data stays out of its model training. Alongside them, the company added pooled group budgets, so a team can share one token allowance.
The timing follows the numbers Anthropic gave investors this week. Second quarter brought an operating profit, and net revenue retention came in at 900 percent: customers spending in Q3 2025 spent nine times as much in Q3 2026. A customer whose bill grows ninefold gets a visit from finance. These reports hand the engineering manager something to say in that meeting, and TechCrunch ties the launch to a planned IPO.
Outside data points the same way. Jellyfish, which makes tools for software developers, published a report on 27 October covering 276,000 of its customers. Tools running on Anthropic models, Claude Code first among them, were the most used, behind close to one in three code changes proposed for review. Spending on those tools is up 10x since the start of the year and now works out to roughly 2 percent of all engineering cost. “Even if you’re saying 2% is still small, that 10x growth is growth that most CFOs didn’t plan for,” Nicholas Arcolano, Jellyfish’s chief scientist, told TechCrunch.
Competition has not dented any of this yet. OpenAI released its Astra model this summer, and the Valley chatter says Codex is now better. Nikolas Albarran, Jellyfish’s AI product lead, told TechCrunch that Codex barely registered in Jellyfish’s data until Astra arrived, and that Anthropic got there first with a strong enterprise sales effort and a product that worked. In his words, “there’s not a big enough reason to switch.” Anthropic’s adoption numbers have held.
Albarran also described why switching gets harder once a company is inside. All the major labs are pouring money into forward-deployed engineers, consultants who sit inside a customer to get it started. TechCrunch describes the new reports as FDE-lite, a scaled-down way to get some of that embedded support without the consultants. Each coding agent has its own habits, down to how it reads agent.md files, so, as Albarran put it, “that cost to switch is not zero.”
Our read adds one more layer. A dashboard that scores your team’s work also stores the history of how your team works, in one vendor’s format. Moving to a rival means rebuilding the baseline as well as the workflows. TechCrunch’s account does not say how the reports score output, and that is the first thing a finance team will ask.
At roughly 2 percent of engineering cost, Claude Code is a small line today. If spending keeps growing at the pace Jellyfish measured this year, that line gets scrutiny at the next budget cycle, and the manager defending it will be holding one of these reports.
Reported by TechCrunch on 28 October 2026.
