Agent Replacement Ladder
Map how agents progressively replace discovery, research, usage, and execution
- Difficulty
- Advanced
- Time to result
- ~months to results
- Steps
- 6
- Confidence
- 94%
The Agent Replacement Ladder maps a progressive shift from human-operated journeys to agent-operated outcomes. At the first level, an agent consumes advertisements and filters providers. At the second, it researches sources and returns a synthesized answer, removing website visits. At the third, a multimodal agent watches the user's screen and guides software use. At the fourth, the agent learns the workflow, receives authenticated tool access, and performs the work directly. Teams use the ladder to identify which customer touchpoints will disappear next and which capabilities must replace them. The output is an agent-ready journey that exposes reliable information, supports guided interaction, provides controlled tool access, and evaluates success through completed outcomes instead of clicks, visits, or manual feature usage.
Origin
Extracted from Marketing Against The Grain as Kieran Flanagan connected agent-consumed advertising, deep research, multimodal guidance, and tool-enabled execution into a broader go-to-market shift.
Core principles
- 01Agent adoption advances through multiple layers of the customer journey
- 02Each delegated layer removes a traditional human touchpoint
- 03Observation and guidance can become autonomous execution once tools are connected
- 04Companies must increasingly serve the agent as an intermediary or primary user
How to run it
- 1
Map the Human Journey
Document the current sequence from seeing an advertisement through researching, selecting, learning, and operating the product.
Pro tip Include every website visit, comparison, signup, tutorial, and manual action.
Watch out Do not map only the marketing funnel; include the work performed after purchase.
- 2
Delegate Discovery
Assess how an agent could consume advertising, interpret buyer constraints, and select candidates without showing ads to the human.
Pro tip Test realistic prompts containing budgets, requirements, and trade-offs.
Watch out Traditional engagement metrics become less informative when the agent is the audience.
- 3
Delegate Research
Determine whether an agent can examine multiple sources and return a comprehensive, cited answer without requiring direct website visits.
Pro tip Provide clear factual content that can be retrieved and reconciled across sources.
Watch out Citations do not guarantee that the human will click through to the cited website.
- 4
Delegate Product Guidance
Use multimodal assistance to observe the interface and guide the user through tasks while learning the workflow.
Pro tip Focus guidance on the user's desired outcome rather than feature tours.
Watch out Screen observation requires appropriate privacy and permission boundaries.
- 5
Delegate Execution
Give the agent controlled access to connected tools so it can perform the learned workflow on the user's behalf.
Pro tip Use scoped authentication, explicit permissions, and reversible actions.
Watch out Broad tool access without safeguards can turn a usability improvement into an operational risk.
- 6
Redesign Around Outcomes
Replace assumptions about human page views and feature use with measures of whether the agent completed the intended job correctly.
Pro tip Track successful tasks, exceptions, approvals, and recovery paths.
Watch out A decline in human interaction may reflect successful delegation rather than customer disengagement.
In the wild
A marketer wants to add and personalize a website chatbot. A deep-research agent examines dozens of websites and returns a cited tutorial, eliminating manual browsing. A multimodal assistant can then guide the marketer through the relevant software. After learning the workflow and receiving authenticated access to the necessary applications, the agent can perform the configuration itself.
→ The journey moves from human research and operation to agent-led synthesis, guidance, and execution.
Common mistakes
Stopping at Search Disruption
Focusing only on reduced website traffic misses the later stages where agents guide product use and eventually perform the work.
Preserving Obsolete Touchpoints
Requiring visits, clicks, and manual navigation adds friction when the customer's agent can complete the task directly.
Granting Tools Without Boundaries
Autonomous execution needs scoped access, authentication controls, observability, and clear limits on consequential actions.
Is it for you?
Best for
It is best for software companies redesigning marketing, acquisition, onboarding, interfaces, and integrations for agent-mediated use.
Not ideal for
It is not ideal for products where direct human participation is itself the central value proposition.
From the transcript
“Agents are pushing all of the things that we use today in the background because the agent is going to be the user of those…”
“Now we're the agent is consuming the paid advertising ads to do the research, because I've given it a pre-programmed set of script characteristics.”
“It's doing the research organically, it's signing up to use the software, and now it's using the software to actually do the things that I…”
From the episode
AI Agents Will Kill Marketing As We Know It, According to Perplexity's Founder