Half II of II of a visitor collection from AI and innovation professional Bruce MacVarish. The combat over who controls the agent layer has already began, and what founders, traders, and CXOs ought to do about it, mapped to the 5 legal guidelines from Half I.
Key takeaways
- The recap: Half I confirmed that agent adoption is gated by belief, not functionality, and named 5 legal guidelines that decide whether or not an agent earns it: steady verification, id as accountability, context as a mortgage, earned (not granted) belief, and pre-emptive governance.
- The brand new proof: Two extra indicators have landed since. Amazon blocked Meta’s Muse from procuring on its web site, and OpenAI launched its personal competing agent, Dots, constructed on practically equivalent belief structure. The blocking combat isn’t solely a belief story … it’s a second combat over which platforms let a reliable agent by way of the door in any respect. The Dots launch confirms the primary combat is actual: a second main platform independently arrived on the identical 5 legal guidelines.
- The suggestions: This piece sizes the ensuing financial system and closes with particular, checklist-form steerage for founders, traders, and CXOs, with every merchandise run by way of the 5 legal guidelines from Half I.
Market Proof: The Sample Retains Repeating
Half I opened with three unbiased readings converging on belief because the gating variable inside ten days of one another:
“If shoppers can recover from the belief issue, in addition to the safety issue, that integration will enhance dramatically.”
– Eric Sheridan, Goldman Sachs Analysis
Goldman’s learn provides a market-structure level Half I didn’t cowl … agentic commerce marks AI’s transfer from the “platform layer” to the “software layer,” monetized like the remainder of the patron web … adverts for a free tier, subscriptions for an ad-free one.
That stacks a second belief query on the primary: an agent funded by the identical promoting enterprise an individual could not absolutely belief with their information. It additionally provides a capital level. U.S. hyperscalers are on tempo to deploy $1.4 trillion in 2027 capex, dedicated no matter how briskly shopper belief resolves.
September 21: An incumbent attracts a line
Two weeks after Muse launched, Amazon started blocking it from procuring on Amazon.com, citing a failure to establish itself as an agent and issues about credential dealing with that Meta disputes.
The extra telling element was that Amazon selected a contractual block over a lawsuit this time, a month and a half after the Ninth Circuit vacated its personal injunction in opposition to Perplexity’s Comet agent, ruling that it’s the person, not the agent-maker, who “accesses” a web site when an agent acts on instruction.
With that authorized concept weakened, a terms-of-use popup is doing the work a courtroom couldn’t.
September 29: A 3rd platform ships the identical structure
Three weeks after Muse, OpenAI launched Dots at its DevDay convention … always-on private brokers on a devoted GPT-6 Astra mannequin, every operating by itself cloud pc.
The safeguards OpenAI described learn like a guidelines from Half I: an auto-review system checks any motion in opposition to an individual’s personal directions earlier than it occurs, and a separate monitoring system can pause or cease a Dot if it detects a security concern. For organizations, specialist Dots arrive with particular person identities, credentials, and entry, built-in with Microsoft’s Agent 365 governance controls … identity-rooted accountability, arrived at independently by a second main platform in three weeks.
OpenAI additionally excluded the EEA, Switzerland, and the UK from the preliminary launch, the identical regulatory fault line already seen in Half I. The launch adopted, by at some point, OpenAI’s personal apology for an earlier agent accessing an Australian authorities web site with out authorization … a reminder that the case for pre-emptive governance in Half I just isn’t theoretical. Three platforms, three weeks, the identical 5 legal guidelines: at this level the sample is the story.
The Aggregators Get Aggregated
There’s a second combat sitting on high of the belief query from Half I, and Amazon’s transfer is the clearest proof it has began. Belief determines whether or not an individual palms a activity to an agent. This combat determines whether or not the agent is allowed to achieve the shelf as soon as they do.
Ben Thompson’s Aggregation Principle (Stratechery) explains why: companies that mixture demand and personal the client relationship seize most of a market’s worth, whereas suppliers behind them turn out to be interchangeable. Amazon constructed its dominance being precisely that aggregator. A horizontal private agent threatens to run the equivalent playbook one layer up … evaluating Amazon in opposition to Instacart or DoorDash on a single order and routing to whichever wins, turning Amazon into exactly what it fabricated from its personal third-party sellers.
The equivalent dynamic is already taking part in out a stage up the stack, in enterprise software program, and a significant platform CEO has mentioned so on the file. Microsoft’s Satya Nadella has described enterprise functions as little greater than CRUD databases with enterprise logic hooked up and has mentioned Microsoft intends to maneuver that logic “aggressively” into AI brokers that learn and write throughout whichever backend a activity touches … a CRM, an ERP system, e mail, file storage … with out discriminating between them.
If that performs out the best way he describes, the functions beneath turn out to be precisely what this part has simply described taking place to Amazon … a skinny, commoditized floor behind whichever agent tier really holds the enterprise logic. An enterprise that owns its personal information however lets a vendor’s agent maintain that logic hasn’t averted this combat. It has solely moved it to a layer most governance conversations don’t have a look at but, which is precisely the place Half I’s Lease-vs-Personal story picks the argument again up.
The identical collapse cuts each methods for a second cause price stating plainly. A multi-repo agent can also be a single level of inbound publicity: no matter reaches it by way of anybody channel … a request, a message, a contact claiming authority it doesn’t have … inherits standing entry to each system that agent is wired into. The aggregation combat described right here is about who will get to behave on an individual’s behalf. A second, quieter combat is about who will get to even attain the agent that may act in any respect, and a company that wins the primary with out constructing for the second has constructed a much bigger goal, not a safer one.
EXHIBIT 1. How platforms are answering the agent query
| Platform | Stance towards procuring brokers | What that indicators |
| Amazon | Blocks broadly (Meta, Google, OpenAI, Perplexity); builds its personal agent | Defending an aggregator place and a big retail-media advert enterprise |
| Shopify | Opens up, with guardrails: agent checkout runs by way of Store Pay for evaluation | Keen to cede the interface if it retains the cost rail |
| Walmart | Has not blocked procuring brokers to this point | Betting on successful share from a extra open posture |
Supply: Creator’s synthesis, primarily based on reporting from Bloomberg, GeekWire, TechCrunch, and TipRanks, September 2026.
Intuition makes the stakes concrete somewhat than hypothetical. The private agent that went viral on iMessage in August 2026 is already routing greater than $1 billion in annualized transaction quantity, by founder-reported figures, roughly half of it journey bookings positioned by way of whichever airline, lodge, or company’s system the agent reaches on an individual’s behalf. That’s actual demand transferring by way of a channel most incumbents hadn’t constructed a protection for, as a result of the channel was a textual content thread, not an app or an internet site.
DoorDash has answered the best way Amazon did by delivery its personal iMessage agent for text-to-order somewhat than opening its ordering stream to Intuition’s. Different platforms have made the other guess already. Shopify, Instacart, OpenTable, Expedia, Ticketmaster, and Plaid have all signed official integrations, competing for the demand a horizontal agent routes somewhat than walling themselves off from it.
The motivation is similar one which decides each aggregation combat: whoever cooperates wins the demand that whoever blocks turns away. Amazon faces the selection each incumbent aggregator faces as soon as a brand new layer kinds above it … preserve defending the connection it owns, or match the entry its extra open opponents provide. Blocking buys time. It doesn’t resolve the selection.
Amazon will push its personal assistant, however a vertical, single-retailer agent begins at a structural drawback in opposition to a horizontal agent that already holds an individual’s e mail, calendar, and years of context … the identical compounding benefit from Half I’s possession dialogue, now working in opposition to the incumbent. OpenAI’s Dots and Meta’s Muse each begin nearer to that horizontal place already, with ChatGPT alone reporting 1.2 billion weekly customers to construct specialist, identity-scoped brokers on high of. Many firms will wish to turn out to be the aggregation layer for private brokers. Only a few could have the distribution to truly get there.
Sizing the Intelligence Financial system
Put the indicators collectively and a market takes form: $40 billion in shopper AI spend, rising greater than 3x in a single yr nearly completely by way of depth somewhat than attain; two billion world customers, with roughly one in seven payers producing over half of all spending; a hyperscale capital dedication measured within the trillions, made unbiased of how shortly belief resolves; and now, a reside combat over whether or not platforms holding the stock will let a trusted agent attain it in any respect.
EXHIBIT 2. The place worth concentrates on the AI Agent Delegation Ladder
| Rung | Dominant monetization sample |
| Inform | Advert-supported or bundled free tier |
| Suggest | Freemium, upgrading to subscription |
| Draft-and-confirm | Premium subscription plus usage-based overlay |
| Autonomous | Excessive-margin take-rate on accomplished actions |
Supply: Creator’s synthesis, primarily based on Menlo Ventures (2026) and Goldman Sachs Analysis (2026) commentary on agentic-commerce monetization.
OpenAI’s personal pricing for Dots matches this sample precisely: a primary agent bundled free into an current Professional or Enterprise Premium subscription, with further brokers or heavier utilization pushed behind a separate flat price … subscription on the suggest rung, usage-based overlay the second an individual desires extra draft-and-confirm capability.
Modeled ahead 36 months throughout 4 funding eventualities, trust-led funding alone outperforms personalization-led funding alone by roughly eight factors in adoption, and that hole holds throughout each use case examined. Relevance will get folks interested by an agent. Confidence in the way it handles their information and judgment is what will get them to truly hand over the duty.
The Advisable Actions
Three teams have a direct stake in getting this proper.
Every merchandise under maps to a number of of the 5 legal guidelines from Half I: verification, id, context, earned belief, and pre-emptive governance.
For Founders
Run the five-law check by yourself product earlier than anybody else does. Are you able to confirm agent habits repeatedly, not after the very fact? Does each motion hint to an accountable id? Is entry revocable with out a help ticket? Does belief compound with observe file? Are your safeguards pre-emptive, not reactive? Buyers and enterprise consumers will ask; reply it first.
Determine early whether or not you’re constructing an agent or a characteristic. Wiring belief into the structure is a founding resolution, not a compliance layer added later.
Sequence the delegation ladder intentionally. Show the agent at rungs one and two, make the trust-building seen, and let demand for rung three pull itself ahead.
Know which facet of the rent-vs-own fork you’re on. Being the trusted platform and serving to folks personal their very own stack are each reliable methods with completely different economics; choose one intentionally.
If you happen to’re constructing an enterprise agent, perceive precisely what you’re asking for. Multi-repo write entry throughout a buyer’s CRM, ERP, and e mail is similar standing authority Microsoft has described transferring aggressively to seize; refined consumers will more and more ask you to show you don’t maintain it, not simply that you just use it responsibly.
Construct the inbound gate as intentionally because the outbound mandate. A multi-repo agent is a single level of entry to each system it’s wired into; an attacker solely wants to achieve it as soon as, by way of any channel, to inherit all of it. Deal with verifying what reaches your agent as core product, not an afterthought to what your agent is allowed to do.
Don’t construct a method on entry you don’t even have. A authorized win for brokers appearing on a person’s instruction doesn’t assure a platform’s cooperation; design for the incumbents that block you, not simply those that welcome you in.
For Buyers
Make the 5 legal guidelines a diligence guidelines, not a values assertion. Ask to see the mandate mannequin, the audit path, and the revocation mechanism for each legislation in Half I, not only a demo of what the agent can do.
Re-underwrite unit economics round depth, not attain. Adoption barely moved in 2026 whereas spend tripled; the compounding metric is utilization depth per person, not new sign-ups.
Count on focus, not democratization. Quicker progress and better per-user monetization for trust-winning platforms favor fewer class leaders; dimension portfolios accordingly.
Map portfolio publicity throughout the rent-vs-own fork. Trusted-platform performs and owned-infrastructure performs usually are not substitutes for one another.
Ask whether or not a startup’s moat is its information or its logic. An organization that hosts a buyer’s information whereas a vendor’s agent tier holds the enterprise logic has outsourced the dear a part of its personal product; worth that the identical means you’d worth customer-concentration threat.
Underwrite platform-access threat as its personal line merchandise. A horizontal agent’s addressable market depends upon which incumbents select to cooperate … an adversarial variable that has nothing to do with how good the agent is.
For Enterprise CXO Groups
Apply the five-law check to vendor choice, not simply your personal roadmap. Ask any agent vendor the identical 5 questions your personal product ought to have the ability to reply.
Begin on the low rungs intentionally, contained in the group. Inform-and-recommend use circumstances construct organizational belief earlier than you grant draft-and-confirm or autonomous authority over buyer information or cash.
Deal with the audit path as a governance requirement. Insist on legible, exportable information of what any agent did, tried, or was blocked from doing earlier than it touches a regulated course of.
Deal with build-vs-buy as an actual resolution. A rented platform agent is quicker to deploy; an owned, personal stack over your personal information prices extra upfront however removes a vendor out of your belief chain completely.
Determine now whether or not you’re an aggregator or a provider in your personal class. If a horizontal agent may plausibly route demand away from you, an open, agent-friendly posture could win extra share than a defensive one; the reply depends upon whether or not you possibly can realistically turn out to be the aggregation layer your self.
Refuse to let any vendor’s agent maintain standing write authority throughout your methods. Insist that cross-system actions route by way of your personal id and permission layer, scoped per activity and revocable, somewhat than a vendor-held credential; proudly owning your information doesn’t provide you with sovereignty if a vendor’s agent tier holds the logic that decides what will get written the place.
Confirm each inbound request earlier than it reaches the agent, not simply each outbound motion it takes. A multi-repo agent wired into your CRM, ERP, and e mail is a single level of entry to all three; deal with the gate in entrance of it as critically because the mandate behind it.
None of this requires ready for a greater mannequin. It requires constructing the components of the system that allow an individual see, and management, what they’re agreeing to … after which letting the adoption curve, and the financial system forming round it, do the remaining.