A marketer shipped software.
That's exactly what happened at Rakuten. Sales teams, finance teams. Within a week of Claude Code rolling in, non-developer teams started completing technical projects on their own. The changes agentic coding is driving go much deeper than most people realize.
"AI helps developers code" — only half the story
It's true that agentic coding tools boost developer productivity. CRED doubled feature execution speed, and TELUS improved code deployment pace by 30%. But what Anthropic's 2026 Agentic Coding Trends Report really shows runs deeper than that.
Here's the core finding: as software development shifted from "writing code" to "orchestrating agents that write code," non-developers started being able to do the orchestrating too.
Rakuten engineer Yusuke Kaji put it plainly: "If I have 5 tasks, I delegate 4 to Claude Code and focus on just one." As developers stepped back, non-developer teams started stepping into technical work.
4 companies, what the numbers say
The case studies in Anthropic's report reveal a consistent pattern. This isn't just a developer productivity story.
At Zapier, there are more AI agents "working" than there are employees. 800+ agents deployed, and company-wide AI adoption hit 89% — an all-time record. And this isn't just IT's achievement. Marketing teams use automated workflows to create blog posts, social content, and keynote speeches. Designers generate live prototypes during customer interviews.
Rakuten's case is the most striking. Under their "AI-nization strategy," non-developer employees started contributing to technical projects without writing a single line of code. Launch timelines dropped from 24 to 5 days (79% reduction). Critical errors fell 97%. Release cycles went from quarterly to biweekly.
| Company | Core Shift | Key Metric |
|---|---|---|
| Zapier | All employees became agent operators | 89% AI adoption, 800+ agents deployed |
| Rakuten | Non-dev teams complete tech projects directly | 79% faster launches, 97% fewer critical errors |
| TELUS | 57,000 employees all building AI solutions | 500,000 hrs saved, $90M+ cost reduction |
| CRED | Developers moved to "PR reviewer" role | 2× faster execution, 10% more test coverage |
The model CRED is moving toward is especially worth watching. "Developers review pull requests only — Claude Code handles coding and testing." The developer role is shifting from executor to overseer.
So what's left for humans to do?
Here's what Anthropic's report keeps coming back to: "active human judgment remains essential". Agents handle execution. But setting direction, reviewing results, and steering the process — that's still on people.
And this applies across the whole organization, not just developers. TELUS built 13,000+ custom AI solutions, and it wasn't just the IT team doing it. Fifty-seven thousand employees each designed solutions for their own problems.
The new skill for the agent era
More valuable than writing code is the ability to clearly instruct agents on what to build and how. That's not a developer-only skill anymore — and that's exactly the point.
Anthropic's framework for agent collaboration involves three loops: humans define goals and context → agents execute → humans review and adjust direction. Human roles haven't diminished — they've moved to a higher level.
How to get your team started
Looking across all four case studies, a clear entry path emerges.
- Identify your most repetitive tasks first
Test writing, documentation, code review comments — tasks with clear patterns and verifiable outputs are the best starting point for agents. - Start with a small pilot team
Both TELUS and Zapier refined their workflows with a small group before going org-wide. Keep failure costs low while finding the patterns that fit your organization. - Open the door to non-developer teams
The reason Rakuten could roll out to marketing and finance in a week is that they lowered the technical barrier. Give people an interface to participate in technical work without needing to code. - Design your Human-in-the-loop review structure
The more autonomy you give agents, the more important the review process becomes. Design explicit loops where humans verify agent output — like CRED's "PR reviewer model." - Build an agent governance framework
Zapier's 800 agents, TELUS's 13,000 solutions — at scale, you need visibility into what each agent does, with what permissions, and how to track it.
Go Deeper
2026 Agentic Coding Trends Report Anthropic's source report — 8 trends reshaping software development, with full enterprise case studies. resources.anthropic.com
Zapier × Claude Enterprise Inside the workflow of a company running more AI agents than employees, and how they hit 89% org-wide adoption. claude.com
Rakuten × Claude Code How product, sales, marketing, and finance teams all onboarded in one week. 99.9% code accuracy, 97% fewer critical errors. claude.com
TELUS × Claude: Fuel iX Platform Strategy for deploying AI to 57,000 team members and processing 100B tokens per month. claude.com
Building Effective Agents — Anthropic Five core agent architecture patterns, from orchestrator-worker to evaluator-optimizer loops. anthropic.com
CRED × Claude Code The full-agentic execution model where developers only review PRs — and what it takes to get there. claude.com
