The One-Question Test
Pam Didner made a pointed observation in Inc. on September 26, 2026: "One concrete question reveals whether the product chooses its own path or simply runs predefined instructions." That question is: when faced with an unexpected situation, does the tool decide what to do next, or does it follow a path you already defined?
This is the simplest way to separate genuine AI agents from automated workflows with a more expensive label. If the tool can only execute what you built, it is automation. If the tool can reason about a new situation and take an action you did not explicitly program, it is an agent.
Both are useful. Both have a place in a marketing technology stack. But they have very different price tags, very different risk profiles, and very different capabilities. Confusing one for the other is an expensive mistake.
Why the Confusion Exists
By late 2026, nearly every marketing tool has added "AI" to its marketing. According to data cited by RTInsights on September 25, 2026, 94% of organizations are already using AI to prepare or execute marketing in some form. A Content Marketing Institute report covered previously showed that 95% of B2B marketers use AI but only 39% see results. The gap between adoption and results is partly explained by tool confusion: teams buy "AI agents" that are actually scripted workflows, expect autonomous performance, are disappointed, and conclude that AI doesn't work.
The terminology problem is real. The word "agent" has become a marketing term, not a technical one. Vendors use it to justify higher pricing and to signal alignment with the current trend. Understanding what is actually under the hood matters more than the label on the box.
Automation vs. Agent: A Functional Framework
The one-question test is a good starting point. For a more complete evaluation, use this five-point framework adapted from the original Inc. observation and informed by work at Vivid on agent architecture.
1. Decision-Making
| Capability | Automation | AI Agent |
|---|---|---|
| Follows predefined rules | Yes | Yes |
| Handles unexpected inputs | No, fails or skips | Yes, adapts |
| Chooses between multiple paths | No, one path per trigger | Yes, selects based on context |
| Improves over time | No | Can, with feedback |
2. Scope of Action
Automation executes a specific task: when X happens, do Y. An agent works toward a goal: when X happens, figure out the best path to Y. For example, an automated email tool sends a follow-up email three days after a contact downloads a whitepaper. An agent reads the contact's engagement with the whitepaper, determines that they downloaded but did not open it, and decides to send a different follow-up focused on the section they spent the most time on instead of the generic three-day email.
3. Error Handling
When an automation encounters an input it was not designed for, it typically fails, skips the item, or sends an error notification. When an agent encounters the same input, it should be able to recognize the unexpected situation, choose a reasonable fallback action, and log what happened for human review. This is the capability that makes agents suitable for unstructured work and automation suitable for structured work.
4. Governance Requirements
Automation needs basic monitoring: did the workflow run? Did it succeed or fail? An agent needs deeper governance: what did it decide? Why did it decide that? What data did it access? What would it have done if the situation had been slightly different? If the vendor cannot answer these questions about its "agent," you are probably looking at automation with agent branding.
5. Pricing Indicators
If a tool costs significantly more when labeled an "agent" than an equivalent tool labeled "automated workflow," but cannot demonstrate decision-making or adaptive behavior, the premium is paying for marketing, not capability. Real agents cost more because they require more infrastructure, more compute, and more governance. Automation costs less because it is deterministic. Paying agent prices for automation is the most common waste in marketing technology budgets in 2026.
Practical Steps for Evaluating Marketing Tools
Step 1: Ask the One Question
Before any demo or trial, ask the vendor's sales representative: "When the tool encounters a situation you didn't explicitly program for, does it choose what to do, or does it follow the closest predefined path?" The answer, and how confident the rep is in giving it, tells you a lot.
Step 2: Test with Edge Cases
During a trial, deliberately feed the tool inputs that fall outside normal parameters. An automation will fail or ignore them. An agent will attempt to handle them, even if imperfectly. The quality of the handling matters less than the fact that it tries.
Step 3: Review the Audit Trail
A genuine agent produces a decision log: what it observed, what it considered, what it chose, and why. If the tool only shows inputs and outputs with no record of decision-making, it is automation. The audit trail is both a governance requirement and a diagnostic signal.
Step 4: Check the Integration Model
Automation connects to specific APIs and triggers. Agents may connect to the same APIs, but they also use broader tools (MCP, function calling, browsing) that let them interact with systems in ways not pre-configured. If the tool says "agent" but only connects to a fixed list of pre-built integrations with no capacity for ad-hoc interaction, be skeptical.
Step 5: Evaluate the Fallback
What happens when the agent is wrong? The best agent tools have human-in-the-loop checkpoints, confidence thresholds for autonomous action, and clear escalation paths. If the vendor says "it just works" without describing what happens when it doesn't, that is a red flag for both automation and agent tools.
Conclusion
Read next:
- Amazon's Agentic Double Move: Opening Seller Central to AI While Blocking Meta's Muse
- Microsoft Autopilot: The 24/7 Cloud Agent That Works While You Sleep
- The AI Shelf: Why Brands Now Optimize for ChatGPT and Gemini
- AI Agents vs Traditional Automation: Business Decision Guide
Evaluating AI marketing tools for your business? Contact Vivid to talk through what you actually need.
