Signal Watch is the living layer of Polaris Shift. It connects important AI developments to behavior, economics, competition, and the strategic questions leaders may need to revisit.Read the latest evidence review →
01EVENT
What happened?
02EVIDENCE
How strong is the case?
03SHIFT
What could change because of it?
04EXPOSURE
Who is most affected?
05STAR QUESTION
What should leaders reconsider?
EDITORIAL DISCIPLINE
A product announcement establishes what was announced. Availability and reliability need separate checks. It cannot establish broad adoption, economic impact, or a strategic shift. Signal Watch keeps those claims separate.
Evidence review · Reviewed October 31, 2026
When the work changes, does the measure still hold?
Before asking whether faster work creates more value, ask whether you are still comparing the same work.
What the source reports
In a February 24, 2026 update, METR said its follow-up developer experiment could not reliably establish the current productivity effect of AI. Some developers avoided participation or withheld tasks because they did not want to work without AI. Lower compensation also affected selection. Concurrent agent use complicated time reporting.
METR considered greater speedup plausible, but described the evidence for its size as weak. It was redesigning the experiment. These are limitations of that study, not evidence that AI cannot improve productivity.
Our speed-versus-value review examines a separate May survey. The February experiment and May survey use different designs and participants. They are not a before-and-after comparison or interchangeable estimates. This review revisits published evidence; it is not an October research announcement.
The Polaris interpretation
A stable dashboard can hide a changing denominator. AI may influence which work teams attempt, which tasks they abandon, and what quality they consider acceptable. Comparing completion time alone could then miss a change in the work itself. This is an interpretation to investigate, not an observed result for your organization.
At the strategic level, ask whether the portfolio still serves the intended outcome. A team could become more ambitious without needing a new North Star. It could also become exceptionally productive at work customers no longer need. The study does not decide between those possibilities.
A test before the next investment
Choose one decision and record two views: comparable tasks with quality and rework accounted for, and the wider portfolio, including newly feasible work, work dropped, and who benefits. Name an owner and review point. Keep elapsed time, human effort, and outcome evidence separate. This is a proposed exercise, not a validated assessment.
What would change this interpretation?
If task selection and quality remain comparable and independent outcome evidence supports the dashboard, the existing measure may still be useful. If the mix changes, revise the comparison before drawing conclusions about the goal.
The Star question: Has performance changed, or has the work we are measuring changed?
Evidence review · Reviewed October 28, 2026 · Updated November 17, 2026
Faster work. More value?
A fuller dashboard can still leave an uncomfortable question: did the work become more useful?
What the source reports
On May 11, 2026, METR published a survey of 349 technical workers conducted in February–April. Respondents reported a median speed multiplier of 3×. Three differently worded questions about work value produced median multipliers ranging from 1.4× to 2×. That range spans the three measures; it is not a confidence interval.
METR asks about work that respondents, their teams, or their leadership would consider valuable. These are respondents’ judgments about their contribution. The survey does not independently measure customer outcomes or business returns.
Read the source methodology ↗ alongside the headline. Use its definition when discussing the estimates; treat our customer-resolution example below as a separate proposed test.
What it does not establish
These are self-reported estimates from a convenience sample, with potential selection bias. They are not measured causal gains, a representative estimate for every workforce, or validation of Polaris Shift. METR warns that respondents may overestimate the effects.
For a different study design and its limits, see the February developer-experiment review. Its estimates cannot be read as an earlier point in the same series.
A measured counterpoint
Published online on November 6, 2026, a study by Lu Fang and colleagues reports randomized field experiments across seven workflows at one cross-border online retailer. The experiments ran in 2023–2024 and involved millions of users and products. Sales effects ranged from no detectable impact to 16.3%, mainly through higher conversion rather than larger carts. The authors report that product return rates and customer ratings did not deteriorate.
This update uses the publisher’s abstract and publication record. The full methods and appendices have not been audited here. These are short-term results from one platform, rather than a universal AI multiplier. Unchanged returns and ratings provide useful guardrails; they do not establish that every shopper found a better fit. The retail experiments and METR survey study different populations, tasks, and measures, so their estimates cannot be combined or treated as a before-and-after series.
The Polaris interpretation
Speed and value deserve separate questions. A team could produce more work while leaving its most important customer problem unresolved. It could also improve the current system in ways that serve both the business and its customers. Test which case you are in. When relevant outcomes improve and guardrails hold, the evidence may support Stay or Adjust before a new North Star.
What would change this view?
Repeated evidence that faster delivery also improves the intended outcome, with quality, rework, and coordination costs accounted for, would support keeping the current direction. A growing gap would warrant examining the workflow, measures, and assumptions before reconsidering the goal.
A small test for your team
Choose one AI-assisted workflow. Record delivery time alongside one outcome people care about, such as reliable resolution or accepted work without rework. Compare like-for-like work over a defined period and record uncertainty. This is a proposed exercise, not a result from the survey.
The Star question: If we finish twice as much work, who is better off, and what evidence would show it?
These entries begin with dated primary-source evidence. Polaris Shift separates what was announced from what it may mean, and keeps the interpretation provisional until behavior and outcomes catch up.
SIGNAL / 01
OBSERVED EVENT
Google previewed Universal Cart in May and reported starting its U.S. rollout in October.
Events: May 19 and 20 and October 15, 2026. Reviewed: November 13, 2026. Google reported a U.S. Universal Cart rollout with in-cart checkout or transfer to a merchant’s site. Its Cart API guide, updated October 16, describes a one-way handoff into the merchant’s cart. Its analytics guide, updated October 27, counts cart transfers as clicks and native purchases as completed orders on Google. It says the report cannot follow activity after a cart transfers to the merchant’s site. These documents provide no merchant results on repeat use or product fit.
Shopping coordination could move into an intermediary while the merchant still completes the purchase. Testing that possibility means following the journey through to a completed order and customer fit. Transfer counts alone leave that question open.
WHAT WOULD MAKE THIS STRUCTURAL?
Compare native orders and transferred carts separately. For transferred carts, use the merchant’s own completed-order records, then examine repeat use, product fit, cancellations, and avoidable returns across journeys. Compare similar shoppers and periods before changing the North Star.
WHAT WOULD WEAKEN THIS VIEW?
More transfers without more suitable completed purchases would weaken this view, as would low repeat use or better outcomes through familiar merchant-site journeys. If the benefit mainly helps shoppers organize a list while the intended outcome stays the same, Adjust may be enough.
EXPOSURE
Retailers, marketplaces, payments, search, digital marketing, loyalty.
STAR QUESTION
If checkout and shopping coordination move into AI-mediated surfaces, should owned-site conversion remain the primary measure of digital commerce success?
Primary-source announcements establish what an organization announced. Availability and reliability need their own checks. They do not establish broad adoption, economic impact, or a structural shift. Those require additional behavioral and outcome evidence.
Strategic lenses
Look beyond the individual announcement.
The current briefs begin with sourced events. The lenses below help interpret recurring patterns across customer behavior, interfaces, work, and economics without pretending that any single event proves a structural shift.
Signal Pulse
Not every headline is a signal.
Polaris Shift looks for developments that may change behavior, economics, interfaces, operating models, or competition. Select a lens to see what deserves strategic attention.
SELECTED SIGNAL / 01
DIRECTION
The journey can compress around an intelligent intermediary.
EXPOSURE
Retailers, marketplaces, payments, search, digital marketing.
WHAT WOULD MAKE THIS STRUCTURAL?
Meaningful customer adoption, reliable transaction completion, and durable changes in traffic or acquisition patterns.
EVIDENCE STATUSPattern emerging
CONFIDENCEModerate
Event
Pattern
Behavior
Structure
EVIDENCE STILL NEEDED
Repeated evidence of customer delegation at meaningful transaction volume and sustained changes in brand-owned traffic.
WHAT WOULD WEAKEN THIS VIEW?
If agent-led shopping remains a narrow discovery aid while customers continue to complete most high-value decisions and transactions through traditional interfaces.
Illustrative strategy lens. These examples generate questions; they are not findings about your organization.
Industry Lens
The same signal can move different North Stars.
Strategic consequence depends on context. Choose an industry to see how one structural signal can create a different exposure, metric tension, and leadership question.
SHARED SIGNALIntelligent agents increasingly act on behalf of customers.
STRATEGIC EXPOSURE
The journey moves outside the site.
Discovery, evaluation, and transaction can increasingly happen through an intermediary that customers trust to act for them.
METRIC UNDER PRESSURE
Site conversion
A site-based metric can become less complete if meaningful customer value is created or captured elsewhere.
STAR QUESTION
If the customer delegates the journey, what outcome should digital commerce optimize for?
POSSIBLE DIRECTION TO INVESTIGATEFrom optimizing a destination to optimizing successful demand capture across interfaces.This is a hypothesis to investigate, not a prescribed North Star.
Illustrative strategy examples. Polaris Shift uses industry context to generate questions, not universal prescriptions.
Pattern view
Watch where pressure is building.
A single event is rarely enough to move a North Star. The strategic signal becomes stronger when multiple developments point in the same direction.
Signal Radar
A signal matters when external change meets an internal contradiction.
The radar separates technological novelty from strategic consequence. Select a signal to see the potential shift, the inherited assumption it puts under pressure, and the evidence that would make the signal more structural.
Prototype visualization. A strong signal is not simply new, large, or fast. It earns strategic weight when multiple forms of evidence show that a material assumption is becoming less true.
The Polaris Brief
Three signals. One shift. One question.
POLARIS SHIFTTHE POLARIS BRIEF / PREVIEW ISSUE
FOR LEADERS NAVIGATING AI-DRIVEN CHANGE
THE IDEA IN ONE SENTENCE
The interface is becoming less important than the outcome it helps an intelligent system complete.
SIGNAL 01
Agents move closer to transaction.
The strategic pressure is not simply another checkout channel. It is the possibility that customer intent, comparison, and action increasingly happen before the traditional site journey begins.
SIGNAL 02
AI becomes a discovery layer.
Visibility begins to include whether intelligent systems can understand, trust, and select an organization's products, knowledge, and capabilities.
SIGNAL 03
Work shifts from assistance to coordination.
As agents span multiple tools and tasks, the useful unit of productivity may become the completed outcome rather than the speed of any one human step.
THE SHIFT TO WATCH
Journey ownership may migrate.
The organization may no longer control every interface through which value is discovered, evaluated, or created.
If that pattern becomes structural, experience strategy changes from designing every step of a controlled journey toward making the organization useful, legible, trustworthy, and executable across a network of human and machine interfaces.
THE STAR QUESTION
Which of your most important metrics assumes that the customer or employee still follows a journey AI may be collapsing?
WHAT WOULD CHANGE OUR MIND?
We would look for durable adoption, reliable agent execution, measurable changes in traffic or workflow patterns, and evidence that value actually migrated rather than simply gaining a new interface.