The Numbers Don't Lie
When we talk about AI agents, the conversation often stays abstract — autonomous systems, tool use, multi-step reasoning. But a critical question rarely gets a precise answer: where are these agents actually being used?
Recent industry data gives us the clearest picture yet. Across thousands of agent deployments tracked in early 2026, one domain crushes every other by an enormous margin.
That's not a slight lead — it's a near-majority. Software engineering alone accounts for more agent deployments than all other categories combined. But the full breakdown reveals a much richer picture.
Why Software Engineering Dominates
The reason software engineering sits at nearly 50% isn't surprising — it's structural. Software engineers are the ones building and deploying these agents. They experience the value first-hand and understand the technical requirements for integration.
More importantly, software engineering is uniquely suited to agentic AI. The work involves clear inputs and outputs, version-controlled environments, automated testing pipelines, and well-documented APIs. Agents can write code, run tests, debug errors, review pull requests, and deploy changes — all within systems that were already designed for machine-readable workflows.
Tools like Claude Code, GitHub Copilot Workspace, and Cursor have accelerated this. A developer today can describe a feature in natural language and watch an agent scaffold the implementation, write tests, and open a PR. The feedback loop is fast and measurable.
The 9.1% That Matters Most
Back-office automation sitting at 9.1% is arguably the most significant number on this chart. Unlike software engineering, back-office work isn't a domain of early adopters — it's the domain of administrative staff, HR teams, finance departments, and operations managers.
The fact that nearly 1 in 10 agent deployments targets these workflows signals a shift: AI agents are moving from developer tools to business operations tools. Tasks like invoice processing, data entry, report generation, email triage, and compliance checking are being automated with agents that can understand context, make decisions, and interact with multiple systems.
The Long Tail: Where the Opportunities Are
Everything below 5% on this chart represents massive, underserved markets. Consider the numbers:
- Marketing & copywriting (4.4%) — agents generating ad copy, social posts, and campaign briefs
- Sales & CRM (4.3%) — lead scoring, follow-up emails, pipeline management
- Finance & accounting (4.0%) — reconciliation, anomaly detection, regulatory reporting
- Data analysis & BI (3.5%) — querying datasets, building dashboards, surfacing insights
- Cybersecurity (2.4%) — threat analysis, log monitoring, incident response
These aren't niche applications. They represent core business functions in every organisation. The low percentages don't mean low potential — they mean low maturity. The infrastructure for agent deployment in these domains is still being built.
The Sub-2% Frontier
The bottom of the chart is where things get interesting for long-term positioning:
- Customer service (2.2%) — despite being the poster child for chatbots, true agentic customer service (where agents take actions, not just answer questions) is still rare
- Healthcare (1.0%) — regulatory barriers slow adoption, but the potential for medical record analysis, scheduling, and triage is enormous
- Legal (0.9%) — contract review, due diligence, case research — the legal industry is data-rich and process-heavy, yet barely touched
- Education (1.8%) — personalised tutoring, curriculum design, administrative automation
What This Data Tells Us About the Market
1. The developer-first phase is peaking
At nearly 50%, software engineering can't grow much further in share. The next phase of growth will come from non-engineering domains catching up. Organisations that position themselves to serve these verticals — with industry-specific agent templates, integrations, and workflows — will capture the next wave.
2. Infrastructure determines adoption
The domains with higher adoption (software engineering, back-office) share a common trait: they have structured, API-accessible systems. Domains with lower adoption (healthcare, legal) tend to have fragmented, legacy systems. The bottleneck isn't AI capability — it's data accessibility.
This is exactly why protocols like Anthropic's Model Context Protocol (MCP) matter. They standardise how agents connect to systems, lowering the integration barrier for every industry.
3. Regulated industries are next
Healthcare, legal, and finance sit at the bottom not because agents can't help — but because compliance requirements create friction. As governance frameworks mature and agents gain audit trails, these sectors will see explosive growth.
Implications for Organisations
If you're not in software engineering, this data is actually good news. It means your industry hasn't been saturated with agent tooling yet. Here's the playbook:
- Audit your data infrastructure — agents need structured, accessible data. If your systems are siloed, start there.
- Start with back-office — it's proven (9.1% and growing) and low-risk. Invoice processing, reporting, and data entry are safe bets.
- Invest in agent-ready APIs — every system your team uses should be accessible via well-documented APIs. This is the foundation for agent deployment.
- Build governance early — don't wait for regulations. Implement approval workflows, audit trails, and human-in-the-loop checkpoints now.
The 49.7% concentration in software engineering isn't a sign that agents only work for developers — it's a sign that every other industry is about to have its agent moment.
What's Next
By the end of 2026, we expect the software engineering share to drop below 40% — not because engineering adoption declines, but because other sectors are accelerating fast. Back-office, sales, and finance will grow the most as enterprise-grade agent platforms mature and MCP-style protocols eliminate integration friction.
The question for every organisation is simple: will you adopt agents before your competitors, or after?
