Agency Guide: Claude Leads Agent Usage at 38%, Creative Work Just 3% of Agent Value

Digital Agency Network (DAN), a London-based platform connecting brands with vetted agencies, published a free report called The Agentic Workflows Guide on September 14. Built with contributions from 10 agency leaders across its member network, the guide documents how agencies are deploying AI agents inside live client work and puts concrete numbers behind the results.

The most direct examples are two rewritten processes. One contributing agency compressed a market intelligence process that previously took more than a week into a single overnight run, while another cut SEO planning and audit projects from one or two days down to a few hours. The guide attributes those gains to restructuring the workflow itself, not simply to running a faster model.

Architecturally, the agencies run both sequential multi-agent systems and parallel orchestration, assigning specialised agents to research, analysis, evaluation, content generation and quality assurance before outputs converge for human review. Agents do the execution work along the pipeline; judgment and sign-off stay with people.

The value breakdown is the report's most counterintuitive finding. Content production accounts for 28% of reported value, followed by data analysis and reporting at 24% and internal operations at 17%. SEO and organic growth and paid media each account for 11%, and personalisation and segmentation 7%. Creative production came last at 3%, a direct counterpoint to the widespread expectation that generative AI would reshape creative work first.

On tooling, agencies are assembling layered stacks rather than committing to one vendor. Anthropic's Claude leads at 38% of reported usage, ahead of OpenAI at 23% and Google's Gemini at 19%. Automation platform n8n accounts for 12% and custom-built internal systems 8%, a spread that suggests tooling is assembled task by task, with different systems handling research, reasoning, automation and quality assurance.

Speed has not come at the cost of oversight. Most contributing agencies run human-controlled or semi-autonomous workflows, keeping strategic review, quality assurance and final approval in human hands. "The architecture matters less than where you put the human," said Adam Griffith, managing director at Luminary. "Speed from AI, judgment from humans. No default model outputs reach our clients," said Michael Gaizutis, CEO and founder at RNO1.

The obstacles agencies name are equally specific: output quality inconsistency is cited most often, followed by a lack of control and predictability and the risk of over-automation. Deployment maturity varies widely. Half of the agencies surveyed describe their implementation as project-specific and 30% call it experimental, while only 10% have standardised agentic workflows across the organisation and 10% operate them at enterprise grade.

The guide closes on what automation does to the agency business model. Contributors point to emerging revenue built on strategic infrastructure consulting, IP-driven monetisation, agent curation and SaaS-like scalability. Jill Maldonado, director of brand growth at Victorious, argues that agentic workflows will pressure the traditional agency model by reducing the value of execution while increasing the value of judgment, strategy and orchestration. The guide is available as a free download from DAN's site.