How DeepStack settles a question · Part 09

Can agents break the habit?

How fast do habits break?

From Part 09: Habit Was the Moat. Agents Are Draining It.. 13 voices on 6 sides, attributed and dated; the DeepStack Call with its history; 5 dated tests that settle it; the results as they come in.

The mechanism

Meta's Muse has inserted itself between shoppers and sellers, and its economic position is what changes the bill. Muse is not merely answering; it chooses, pays and decides which seller is seen. A business paid by customer inertia, a forgotten subscription, a commodity brand, an ad-funded marketplace, loses the moment an agent makes comparison free, while trust, regulation, system-of-record status, distribution and compute remain defensible because the agent needs them to act. Paid consumer AI adoption is still about 3 percent, so the stock moves of September, Meta up 18 percent and AMD up 33 percent, are an early invoice rather than a finished income statement. The Call therefore backs the rails over the habit, and would change if Meta reported downloads without retained users, transaction volume or fees.

Approved by Marco Cauduro on October 10, 2026.

The DeepStack callSplit

Agents will drain weak consumer habits first. Back the rails (trust, records, distribution, compute), not the forgotten subscription.

Sell habit, buy the rails · Conviction Moderate (3/5)

Our Call takes the side “Eighteen months, or less”.

Read the full call

Call history2 versions · last changed
  1. v2Sell habit, buy the railsConviction Moderate (3/5)Side: Eighteen months, or lessSide named by the owner.
  2. v1Sell habit, buy the railsConviction Moderate (3/5)First call, as published with the story. The conviction, the actions and the flips were set down for the site on October 7, 2026.

6 voices

Eighteen months, or less

  1. Andrew Walker

    Faced with two extreme camps on consumer AI agents (that they are massively overhyped versus that every consumer-facing company is a short because agents will eat their margins), Andrew Walker says both are too extreme but if forced to pick, they…

    Source

  2. Andrew Walker

    AI agents make price-shopping and replacing skilled workers (e.g., using AI to edit a podcast instead of a consultant) easy, and if saving money is the core consumer use case for AI, this could have a deflationary impact on the consumer economy at…

    Source

  3. Tae Kim

    As computer tool use rises, AI compute demand will explode just as it did when agentic coding took off, when reasoning models ramped, and when AI chatbots were first released.

  4. Tae Kim

    Having used Muse over the past week, Tae Kim found problems in the quality and accuracy of its responses and found it hard to choose it over ChatGPT, and suspects others will feel the same way.

    Source

  5. Ren

    Once an AI moves from surfacing a link to researching, deciding, executing and paying, it inserts itself between the customer and the business, potentially capturing an unusually valuable piece of the internet's discovery and transaction funnel.

    Source

  6. Ren

    The unit of interaction is shifting from the prompt to the objective: instead of a chatbot loop of question-and-answer, Muse takes a goal, plans, executes, and only interrupts the user at decisions it isn't allowed to make alone.

    Source

6 voices

Still a demo

  1. Moses Sternstein

    The data suggests that, for now, the story is 'App-Slop': AI has produced many more apps, but they are not successful apps in any meaningful sense.

    Source

  2. Moses Sternstein

    Paid consumer AI adoption remains quite low even though it is growing, per Consumer Edge data showing roughly 3% adoption as of Q1, with younger people adopting at about 4x the rate of older people.

    Source

  3. David George

    The creators argue OpenAI will prevail not because of best models, chips, cost-performance curve or even best product — though they think OpenAI has those — but because it is so good at creating new kinds of customers and has the most durable…

    Source

  4. David George

    As of September 2026, the creators say intelligence-as-a-service product rankings are no longer the most important question; intelligence is everywhere and nearly anyone can do nearly anything, yet most people aren't yet, so awakening new behavior…

    Source

  5. Michael Spencer

    Michael Spencer is not confident about Muse's odds of success, citing that only Anthropic has shown how to scale ARR so far in this Generative AI cycle, that the ratio of app downloads to actual users seems fairly low and suspicious, and that Meta…

    Source

  6. Michael Spencer

    Generative AI's impact on the macro economy and labor market is a serious concern: higher inflation, less hiring, more anxiety, and a lack of ROI compared to the debt, capex and resulting AI bubble, with higher earnings masking dangerous…

    Source

3 voices

Voices that cross sides

  1. Andrew Walker

    Andrew Walker says they cannot watch how easy Muse is to use and how fast its downloads have taken off and conclude it won't have a big impact on the consumer economy.

    Source

  2. Andrew Walker

    Andrew Walker suggests that Muse acting as a layer that decides where consumer spend goes is a massive threat and opportunity for retailers, noting as evidence that Amazon banned Muse.

    Source

  3. Tae Kim

    On the owner of the agent and on the rails beneath it.

    Source

Where they cross

Who answers whom.

The specific points where one voice meets another: the same evidence read two ways, or the same mechanism with a different sign.

Alternative explanations

Three other readings.

What would have to be true for the Call to be wrong, as the other side argues it.

  1. The habit is not breaking yet: AI has produced many more apps and few successful ones, and paid consumer adoption sits near 3 percent, growing but still small (the case Moses Sternstein makes).
  2. The agent is not good enough to be chosen: a week of using Muse turned up quality and accuracy problems and no reason to prefer it over ChatGPT, and others will feel the same (the case Tae Kim makes).
  3. The dramatic prediction is usually wrong: radiology was to be taken over by AI within five years and instead more radiologists are needed, so the incumbents keep more than the Call allows (the case Jensen Huang makes).

What settles it

The dated tests.

The tests the story set, each with its specification where the owner has written one and the piece of the argument it verifies; on the Docket, resolved on the Results page.

  1. Meta reports its September quarter; users, transaction volume and fees would show whether downloads are becoming a business.

    Part 09: Habit Was the Moat. Agents Are Draining It. Pendingp09-t01Pre-specifiedChecks the retained-users claim

    Question
    Did Meta's September-quarter release, slides or earnings call state a number of users, a transaction volume or a fee revenue for Muse, beyond download or install counts?
    Source
    Meta investor relations, third-quarter 2026 results
    Rule
    Read the release, the slides and the call transcript on Meta's investor-relations site. The answer is yes if they state a count of monthly or weekly Muse users, a transaction volume, or a figure for fee revenue or recurring revenue from Muse; no if they give download or install counts alone, or no Muse figure.
    If the source is late, revised or silent
    If Meta reports after November 30, 2026, or restates the quarter, the test is inconclusive, and the Result says what was read and what was missing.

    The Call, before the dateExpects yes. Gives the yes answer 55%.

    55%

    0 · no100 · yes

  2. Booking reports room nights and direct mix; the result tests whether the agent or the incumbent owns the booking.

    Part 09: Habit Was the Moat. Agents Are Draining It. Pendingp09-t02Pre-specifiedChecks the booking-ownership claim

    Question
    Did Booking Holdings report September-quarter room nights above the September quarter of 2025, with a direct-channel share of bookings no lower than the share it reported a year earlier?
    Source
    Booking Holdings investor relations, third-quarter 2026 results
    Rule
    Read room nights and the direct mix from the release and the call. The answer is yes if room nights are above the year-earlier quarter and the direct share is at or above the year-earlier share; no if either declined.
    If the source is late, revised or silent
    If Booking gives no direct-mix figure for the quarter, or reports after November 30, 2026, the test is inconclusive, and the Result says what was read and what was missing.

    The Call, before the dateExpects no. Gives the yes answer 40%.

    40%

    0 · no100 · yes

  3. Nike holds its investor day; margin targets and demand plans show how much selling still assumes human choice.

    Part 09: Habit Was the Moat. Agents Are Draining It. Pendingp09-t03Pre-specifiedChecks the human-choice claim

    Question
    Did Nike, in the materials published for its investor day of November 16 and 17, 2026, state a numerical operating-margin or gross-margin target for a fiscal year beyond fiscal 2027?
    Source
    Nike investor relations, investor day presentation and press release
    Rule
    The answer is yes if the presentation, press release or webcast transcript on Nike's investor-relations site states a margin target with a number and a fiscal year beyond 2027; no otherwise. How much of the demand plan still assumes a person choosing (stores, app, wholesale, marketing) is DeepStack's reading from the same materials and goes in the Result, not in the answer.
    If the source is late, revised or silent
    If the investor day is postponed past November 30, 2026, or the materials are not public, the test is inconclusive, and the Result says what was read and what was missing.

    The Call, before the dateEither answer fits the Call. Gives the yes answer 60%.

    60%

    0 · no100 · yes

  4. The first tax season with agents tests whether households still pay Intuit for certainty or ask an agent.

    Part 09: Habit Was the Moat. Agents Are Draining It. Pendingp09-t04Open questionChecks the certainty-premium claim

    Specification pending

  5. Amazon standoff

    Amazon allowing agent checkout would show the advertising pool beginning to move; continued resistance shows its value remains protected.

    Part 09: Habit Was the Moat. Agents Are Draining It. Pendingp09-t05Open questionChecks the advertising-pool claim

    Specification pending

What changed

The record of Part 09.

The Part’s publication, every change to the Call, every result that came in and the next dated test, oldest first.

  1. PublishedPart 09 published; the Call: Sell habit, buy the rails.
  2. Call updatedThe Call: Sell habit, buy the rails. Agents will drain weak consumer habits first. Back the rails (trust, records, distribution, compute), not the forgotten subscription. Our Call takes the side “Eighteen months, or less”. by Marco Cauduro
  3. Next testNext: Oct 21, 2026, Meta reports its September quarter; users, transaction volume and fees would show whether downloads are becoming a business. Pre-specified
Sources (12)
  1. Reuters - Meta launches Muse
  2. CNBC - Amazon blocks Muse
  3. 24/7 Wall St. - enterprise repricing
  4. Apollo - technology cash-flow question
  5. a16z - AI shopping economics
  6. Yet Another Value Blog - Muse test
  7. Capitalist Letters - Booking analysis
  8. Market Sentiment - software moats
  9. Invest in Assets - Intuit moat
  10. Prof G Markets - Muse market value
  11. The VC Corner - agentic payments
  12. Fiscal.ai - Ennismore software moat

Market data are as of the dates cited. Voices are paraphrased from public writing and attributed by name.

Analysis and opinion, not investment advice. Voices are paraphrased from public writing and attributed by name; DeepStack shows no score, ranking or accuracy for any person. Every question, one page each.