MMarketing Against The Grain
← All frameworks
Productivity

Web-as-Agent-Context Loop

Turn an open web page into context for immediate analysis and modeling

Difficulty
Easy
Time to result
~days to results
Steps
5
Confidence
95%

The Web-as-Agent-Context Loop treats the active web page as a direct input to analysis instead of an endpoint that must be copied elsewhere. First, the user or agent opens a relevant listing, report, or page. Next, the user states the decision to support and asks the agent to extract the necessary variables. The same agent then converts those variables into an artifact such as an ROI model, comparison table, or recommendation. This keeps discovery, extraction, and reasoning in one conversational context and eliminates much of the copying into spreadsheets or separate chat windows. Verification remains essential: users should confirm prices, dates, and assumptions against the original page before acting on the model.

Origin

Extracted from Marketing Against The Grain during a discussion of analyzing a property directly from its browser context.

Core principles

  • 01Treat the current page as structured working context
  • 02Keep extraction and analysis in one conversational loop
  • 03Ask for a decision artifact rather than a generic summary
  • 04Reduce manual copying between browsers, spreadsheets, and models

How to run it

  1. 1

    Load the source page

    Navigate to the listing, report, or web resource containing the facts needed for analysis.

    Pro tip Have the agent locate the page if discovery is part of the task.

    Watch out Confirm that the page is the correct and current source.

  2. 2

    Frame the decision

    State what you are deciding and which outcome matters, such as payback period or expected return.

    Pro tip Specify the units and time horizon.

    Watch out A generic request may produce a generic summary instead of a useful model.

  3. 3

    Extract the inputs

    Ask the agent to identify the relevant data from the page and distinguish facts from assumptions.

    Pro tip Request a visible input table before calculations.

    Watch out Web pages may omit costs or contain promotional estimates.

  4. 4

    Build the artifact

    Have the agent create the requested model, comparison, or decision analysis within the same context.

    Pro tip Request best-case, base-case, and worst-case scenarios.

    Watch out Do not treat inferred values as source facts.

  5. 5

    Verify the model

    Check critical inputs against the page and revise assumptions before making a decision.

    Pro tip Focus verification on variables that most affect the result.

    Watch out Convenient context transfer does not guarantee analytical accuracy.

In the wild

Campsite ROI model

After opening a low-cost land listing, an investor asks the browser agent to use the property's price and characteristics to model campsite revenue and estimate the payback period. The analysis happens without manually moving the listing data into a spreadsheet or separate chatbot.

The investor receives a decision-oriented model directly from the active listing.

Common mistakes

Skipping source verification

The agent may misread a page or silently infer missing values, so decisive inputs must be checked.

Requesting analysis without a decision

Without a clear decision objective, the agent may summarize the page rather than produce a useful artifact.

Is it for you?

Best for

Users who routinely turn listings, reports, or product pages into calculations and decisions.

Not ideal for

High-stakes analysis where source data cannot be independently verified.

From the transcript

But now it's just context in a window.

Kieran Flanagan · 08:30

So you can just say to the AI, okay, like take this and then I build like a ROI model of how I can extract…

Kieran Flanagan · 08:30

And so it turns the web into like context for your agent.

Kieran Flanagan · 08:30

From the episode

OpenAI Just Launched a Web Browser (and It’s Smarter Than Chrome)