Most 'AI Agents' Are Just Workflows — the Test

Here is the uncomfortable truth: most products marketed as "AI agents" are multi-step workflows with an LLM doing text processing in the middle. The spectrum from a bash script to a true autonomous agent is real, but the marketing collapses it into a single buzzword.

A script follows fixed rules. An automation adds triggers. A workflow chains multiple steps. An AI workflow adds LLM reasoning at one or more steps. A genuine AI agent sets its own sub-goals, selects its own tools, and operates autonomously toward an objective.

Most "agents" on the market today are AI workflows at best — and that is fine, except the pricing and expectations are set at the agent level. The spectrum matters: Script → Automation → Workflow → AI Workflow → AI Agent. Most products sit at level 3 or 4 while marketing themselves as level 5.

Gartner coined the term "agent washing" to describe what is happening: vendors rebranding existing chatbots, RPA bots, and workflow tools as "AI agents" without adding meaningful agentic capabilities. Of thousands of vendors claiming agent capabilities, Gartner estimates only about 130 are real (Gartner, June 2025).

The consequences are measurable. The AI agent market hit $7.63B in 2025 (Grand View Research) — yet Gartner found that less than 1% of enterprise software included agentic AI in 2024, and expects 33% by 2028. It also predicts over 40% of agentic AI projects will be scrapped by the end of 2027 due to escalating costs, unclear business value, or inadequate risk controls.

IT spending on agentic AI is projected to reach $1.3 trillion by 2029 — over 26% of total IT spend (IDC, 2025) — and Gartner now puts $234 billion of enterprise application-software spend at risk of substitution by agents (Gartner, July 2026). The market is massive, but most of the money is chasing marketing, not capability.

The pattern repeats every cycle. In the 2000s we wrote scripts and macros. In the 2010s it became "Robotic Process Automation" — same concept, better branding, a multi-billion-dollar market. By 2018, RPA became "intelligent automation" when vendors added ML classifiers. In 2023, "AI assistants" and "copilots" became the label. By 2025–26, everything is an "AI agent."

Each rebrand adds a genuine capability increment. LLMs did make workflows smarter. But the core value proposition — "software does a task so a human doesn't have to" — has not changed since the first cron job.

The insight is correct: "hire an AI agent" is, for most products, just another way to say "take my automated workflow for a specific task." A well-designed n8n or Make workflow with LLM nodes at key decision points solves 80% of the use cases vendors are selling "agents" for — at a fraction of the cost.

True agentic capabilities — goal-directed, tool-selecting, self-correcting — do exist in products like Claude Code, Devin, and specialized enterprise deployments. The technology is real, just not as widespread as the marketing suggests.

The opportunity is in the gap: build reliable AI workflows while competitors chase the "agent" label. Reliability beats autonomy for most enterprise buyers today. Before you spend six figures on "AI agents," ask one question: does this product set its own goals, or follow a predefined pipeline? The answer saves you six figures.