The global AI seller landscape

Proven rule. Every body claim traces to a named source with an accessed date. Inferred, single-sourced-shaky or contradicted-without-a-winner claims are in §4.

Source legend

All URLs accessed 2026-09-05 unless a different date is stated. Sources are marked P (primary: the vendor's own site, docs, pricing page, press release, or a platform's own documentation) or T (trade/third-party reporting) or X (vendor-adjacent blog — treat as claim, not evidence).

1. What we found

1.1 Headline

The write ceiling from 006 has been broken — but only inside the advertising console, only by paid platforms, and not for a Gulf seller. Between April and August 2026 four serious products shipped something that changes state on Amazon: Pacvue Agent [A1-S10], Profasee Ultra [A1-S19], CommerceIQ's content/bid/promo agents [A1-S17], and Superfuel's title agent [A1-S5]. Every one of them is scoped to a narrow surface (ads, price, or the title field), every one keeps an approval gate or an explicit ramp, and none of them documents support for amazon.ae, amazon.sa or noon.com.

Meanwhile the loudest launch of the period — Helium 10's "Helium", the flagship "Commerce AI Agent that helps run your Amazon business", launched 2026-08-24, five days after 006 closed — cannot write at all. Its own press release: "At launch, Helium is focused on analysis and recommendations; the platform will continue to expand Helium capabilities, including write execution capabilities soon" [A1-S1]. The market leader's agent is an advisor.

1.2 The three classes, and where the line actually falls

• Advisor — chats and recommends; no artifact, no write. Helium [A1-S1], Helium 10's MCP Connector [A1-S3], DataHawk's "Sherlock" [A1-S13].
• Co-pilot — drafts an artifact the human applies. ZonGuru [A1-S23], Jungle Scout Catalyst [A1-S25], AdLabs [A1-S16], Atomic One [A1-S8] [A1-S9], Lumian [A1-S27].
• Operator — writes to the platform. Pacvue Agent [A1-S10], CommerceIQ [A1-S17], Profasee [A1-S19], Superfuel [A1-S5], Scale Insights [A1-S20], Trellis [A1-S18], Perpetua [A1-S11], Teikametrics [A1-S12].

The honest boundary is not advisor-vs-operator, it is what surface the write touches. Every operator in that list writes to ads, price, or one listing field. Not one writes to the surrounding business.

1.3 What "agent", "autopilot" and "end-to-end" actually mean in 2026

Every vendor using the words, and what the words resolve to on their own pages:

Nobody means the business. Every "end-to-end" in the market is end-to-end inside the Amazon console, and in most cases end-to-end inside the ads console. Ilia's SCOPE definition of end-to-end — trade licence → sourcing → freight → customs → VAT → compliance → listing → ads → reorder — has no product pointed at it. See §1.6.

1.4 Amazon now has a written rule for this product class

Effective 2026-03-04, Amazon amended the Business Solutions Agreement and introduced a new Agent Policy [A1-S32]. Amazon's own definition: "'Agent' means any software or service that takes autonomous or semi-autonomous action on behalf of, or at the instruction of, any person or entity" [A1-S32]. Agents must "clearly identify themselves as automated systems", "comply with the new Agent Policy at all times" and "cease access if Amazon requests" [A1-S32]. The same update adds restrictions on using "Amazon materials or services for AI development with enhanced protection against reverse engineering" [A1-S32], and renames the "Developer Site" to the "Solution Provider Portal" [A1-S32].

Trade reporting adds that agents must not simulate human behaviour patterns or bypass CAPTCHAs [A1-S33]. That is a direct constraint on the browser-automation route into Seller Central, which is the only route available to a product without a registered SP-API application. Note that Amazon's own announcement acknowledges the Agent Policy help link (GS83KH2MA7HM69PH) routed to the wrong page [A1-S32]; the policy text itself was not readable from outside a seller login on 2026-09-05 — see §4.

The 006 ceiling ("exports in, recommendations out, human pushes the change" 006:[L18] 006:[L20]) is therefore no longer a technical description — it is now a licensing regime. A product that writes must be a registered SP-API / Ads API application that announces itself as a bot. That is a compliance bar, not a moat, but it is a bar.

1.5 The GCC answer: the pipes exist, nobody points at them

This is the most useful thing in this Area, and it is a two-part finding.

The platform access exists. - SP-API supports both Gulf marketplaces: Saudi Arabia A17E79C6D8DWNP and United Arab Emirates A2VIGQ35RCS4UG, both served from the Europe/Middle East/India/ Africa regional endpoint [A1-S31]. - Amazon runs a Selling Partner Appstore on amazon.ae, advertising "2500+ applications" across "20+ countries" and "1.4MM+ sellers", with categories including automated pricing [A1-S34]. - noon.com operates a public API platform with documentation at noon-docs.noonpartners.dev, explicitly serving "service providers and partners building tooling" and "system integrators", not only sellers, with OAuth, "granular permissions, and auditability" [A1-S29]. Write surfaces named: Content (catalog), Pricing, Stock, Offer, and Fulfilled-by-Partner order/shipment APIs [A1-S29].

The products do not. - ZonGuru states plainly it "currently supports North America, Australia, and European marketplace connections" [A1-S24] — GCC excluded, from the vendor's own pricing page. - Scale Insights lists 12 supported countries (US, CA, MX, UK, DE, FR, IT, ES, NL, JP, SG, AU); UAE and Saudi Arabia are absent [A1-S20]. - Jungle Scout Catalyst is "compatible with marketplaces in the United States, Canada, Mexico, United Kingdom, France, Germany, Italy, and Spain" with unspecified "partial support for additional regions"; Cobalt covers 19 marketplaces [A1-S25]. - SellerApp is the clear exception: 19 marketplaces with "Saudi Arabia" and "UAE" explicitly listed, on every tier including the $0 Freemium [A1-S21]. - Perpetua, Teikametrics, AdLabs, Trellis, Profasee, CommerceIQ, Quartile, DataHawk, Superfuel, Atomic One, Lumian — none names a Gulf marketplace anywhere on the pages read [A1-S11] [A1-S12] [A1-S16] [A1-S18] [A1-S19] [A1-S17] [A1-S15] [A1-S13] [A1-S5] [A1-S6] [A1-S8] [A1-S27]. Silence is not a "no", so these are parked in §4 as unconfirmed rather than asserted as unsupported. - Zero of the ~25 products read mention noon.com. The only software found writing to noon is non-AI multichannel plumbing: Omniful (order sync, inventory push, catalog upload, order cancellation, hub mapping) [A1-S30] and ChannelEngine (marketplace guide exists; page blocked to fetch, see §4). - Amazon Ads' own Middle East guide, aimed at UAE and Saudi sellers, discusses Sponsored Products only and mentions no AI tooling, no API, and no third-party partner tools [A1-S36].

Conclusion: the GCC gap is a go-to-market gap, not an API gap. Anyone can call SP-API for .ae/.sa and noon's partner APIs today. Nobody in the AI-seller category has bothered.

1.6 Pre-product steps: proven "no", with one partial exception

The Brief asked me to prove the "no". Reading every product page above for sourcing, supplier vetting, freight, customs, entity setup:

• Jungle Scout Supplier Database is the single genuine pre-product feature in the whole landscape: search "our database of all global suppliers", verify a known supplier by customer volume or confirmed shipments, reverse-look-up the manufacturer behind any ASIN, plus a Supplier Tracker for contacts and quotes [A1-S26]. It stops at the supplier. Freight, customs, importing and company setup are not addressed on that page [A1-S26].
• Helium 10 has no sourcing, freight, customs or company-setup tool in its tools index; the only Alibaba touchpoint is a co-marketing workshop, not a product [A1-S4]. Its categories are research, listing, inventory, analytics, PPC and reviews [A1-S4].
• ZonGuru claims "official partners with Alibaba and Amazon" for data access [A1-S23], which is a data-sourcing claim, not a sourcing workflow.
• 8fig — the closest thing to a capital-and-supply-chain layer, AI-priced offers in ~24h — requires the business to be "based in the U.S. or Canada", with 6+ months trading, $12K+ average monthly revenue and $100K+ annual revenue [A1-S28]. Structurally unavailable to a Gulf seller, and unavailable to a pre-revenue one anywhere.
• Lumian explicitly does not cover sourcing, freight, customs or company setup [A1-S27].
• Amazon's own seller-tools index lists Seller Central, the Seller app, the Appstore, Automate Pricing, A+ Content, Manage Your Experiments, Brand Analytics and Product Opportunity Explorer — plus an "AI brand name generator", the only AI on the page [A1-S35]. Nothing upstream of "you have a product".

No product read on 2026-09-05 covers trade licence, entity formation, importer-of-record, customs clearance, VAT registration, ESMA/SFDA conformity or Arabic labelling. Not one page mentioned any of them. A7 owns the spine itself; A1's contribution is that the software column of that spine is empty above the "listing" row.

1.7 Is anything priced for a pre-revenue beginner?

Yes — but only the tools that assume you already sell.

And the cost of actually operating:

The pattern: the free and $15–$49 tier is analytics and copy. The moment a product writes to the ads console, the price becomes a percentage of ad spend or a $149–$698 floor. A pre-revenue GCC importer with no ASIN, no ad spend and no orders is not a customer for any of them — several are literally unusable before the first sale, because they price off trailing sales [A1-S6] or require a live ad account [A1-S16].

1.8 AI-native entrants 006 could not have seen

006 closed 2026-08-19. Everything below is dated after, or was missed:

The shape of the 2026 cohort is the story. Two of the loudest "AI-native, autonomous, end-to-end" entrants — Atomic One [A1-S8] and Lumian [A1-S27] — resolve on their own sites to humans approving agent drafts, at agency price points, for brands that already have revenue. The genuinely autonomous writes came from incumbents (Pacvue, CommerceIQ) and from a narrow specialist (Superfuel, one field; Profasee, price and bids).

1.9 Horizontal agents as competitors

006 established the pattern — Claude Code / Cowork briefed with exported files, "more productive than an entire VA team for about $200/month" 006:[L13], an 11-hour research job in 6–7 minutes from one Cerebro export 006:[L20]. Two things changed since:

1. Read access got official. Helium 10's MCP Connector puts live product, keyword, Brand Analytics, advertising and P&L data into Claude and ChatGPT, 30+ tools, and states the data "is never scraped or used for model training" [A1-S3]. Intentwise ships an equivalent "AI Gateway (MCP)" into Claude and Copilot [A1-S14].
2. Write access did not. The Helium 10 connector is described entirely in query and monitoring terms, with no evidence of campaign, bid or listing writes [A1-S3].

So the horizontal-agent stack in September 2026 is: frontier model + vendor MCP for reading, human for writing — with the additional constraint that any write path a builder invents must now register as an Agent and announce itself under Amazon's Agent Policy [A1-S32] [A1-S33].

2. Capability / object table

Columns per the Brief. Write-back? names the mechanism. GCC available? means amazon.ae / amazon.sa / noon.com. Cells are terse by design; the reasoning is in §1 and the caveats in §4.

3. Reconciliation notes

Reconciliation note — Helium 10's agent. Helium 10's marketing frames Helium as "The Commerce AI Agent that helps run your Amazon business" [A1-S1], and its pricing page describes the AI Agent as a partner that "diagnoses problems, surfaces opportunities, recommends optimizations" [A1-S2]. The body of the same press release states: "At launch, Helium is focused on analysis and recommendations; the platform will continue to expand Helium capabilities, including write execution capabilities soon" [A1-S1]. Resolved toward the release body: Helium is an advisor. The headline verb "run" is marketing.

Reconciliation note — Atomic One. The homepage claims an "AI-native operating layer that autonomously manages the end-to-end P&L of your Amazon business" [A1-S8]; the press release says the agents "operate continuously without human oversight" and automate ~80% of repetitive tasks [A1-S7]. The same homepage also states: "Specialist agents draft actions. Brand managers, operators who've scaled real Amazon brands, approve the high-stakes calls" [A1-S8], and the how-it-works page describes SP-API and Ads as read sources on a 60-second interval with operator review [A1-S9]. Resolved toward the workflow description: Atomic One is a co-pilot with humans in the loop, and the presence of named human brand managers suggests a service, not self-serve software. Escalated per the Brief.

Reconciliation note — Amazon Seller Assistant. A vendor blog describes an "always-on agentic AI partner" running on Bedrock (Nova + Claude), free to all third-party sellers, US from December 2025, EU/UK through Q1–Q2 2026, with tiered "suggest only" vs "auto-approve" permissions [A1-S38]. Amazon's own seller-tools index at sell.amazon.com/tools does not list Seller Assistant at all, and the only AI on that page is an "AI brand name generator" [A1-S35]. Not resolved. One source is a competitor's blog; the other is a marketing index that may simply be stale. Parked to §4 and handed to A2, which owns first-party.

Reconciliation note — Helium 10 GCC coverage. Multiple affiliate and review sites state Helium 10 supports amazon.ae and amazon.sa. Helium 10's own knowledge base (kb.helium10.com) and Jungle Scout's own help centre both sit behind a Cloudflare challenge and could not be read on 2026-09-05. Not resolved from primary sources. Parked to §4 — but note that Helium 10 runs podcast episodes specifically on selling in UAE and Saudi Arabia, which is evidence of market attention, not of tool coverage.

Reconciliation note — "AI-native" 2026 entrants. Press coverage frames Atomic One and Lumian as autonomous AI products [A1-S7] [A1-S39]; both vendors' own sites describe human operators in the execution path [A1-S8] [A1-S27]. Resolved toward the vendor sites. The 2026 "AI-native" cohort is, so far, agencies with agents.

4. Open Questions / Parked

Write mechanism, unproven - Superfuel's publish path. It says the agent "updates both fields live" and can "publish on its own" [A1-S5], but names no API. help.superfuel.io does not resolve; /faq 404s; the blog says nothing [A1-S6]. SP-API Listings Items is the obvious inference — not asserted. - Helium 10 Ads, Teikametrics, Perpetua, Trellis, Quartile, Intentwise, sellerboard. All describe automation; none states "Amazon Ads API" on the pages read. Amazon Ads API partner status is the near-certain mechanism — inferred, therefore parked. - Atomic One's write path. SP-API and Ads are named as read sources [A1-S9]; no write mechanism is documented anywhere on the site or in the release [A1-S7] [A1-S8]. - Whether anyone uses browser automation for write-back. No vendor read admitted to it, and Amazon's Agent Policy now makes it hazardous [A1-S32] [A1-S33]. Absence of admission is not absence of practice.

GCC availability, unproven - Perpetua, Teikametrics, AdLabs, Trellis, Profasee, CommerceIQ, Quartile, Intentwise, DataHawk, Superfuel, Atomic One, Lumian do not state whether they support amazon.ae or amazon.sa. Silence on a marketing page is not a "no". Each needs a one-line vendor question before anything is claimed at a partner. - Helium 10 and Jungle Scout feature-level GCC coverage. Both help centres were unreadable (Cloudflare challenge, 403 to both WebFetch and curl). Jungle Scout's pricing page is primary and lists 8 core Catalyst markets excluding GCC [A1-S25]; the widely repeated claim that Product Database, Keyword Scout, Opportunity Finder and Sales Estimator are unavailable for UAE comes only from third-party blogs and a search-engine summary of the help centre — not verified. - ChannelEngine's noon capabilities. Its noon marketplace guide (support.channelengine.com/hc/en-us/articles/16319074295197-noon-marketplace-guide) exists but returned 403 on 2026-09-05; only Omniful's noon docs were readable [A1-S30]. So the claim "ChannelEngine writes to noon" is unverified — Omniful is the proven one.

Platform, unproven - The Agent Policy text itself. Amazon's announcement is primary [A1-S32], but the policy document (GS83KH2MA7HM69PH) was not readable outside a seller login, and Amazon's own post admits the link was mis-routed [A1-S32]. Whether the Agent Policy applies to amazon.ae and amazon.sa BSAs, or only to the US/CA agreement it was announced under, is unknown. A6 should settle this — it determines whether a Gulf-facing operator is regulated by it. - Amazon Ads MCP server status. 006 recorded it as partner-only 006:[L09]. One vendor blog claims an Amazon Ads MCP server launched 2026-02-02 [A1-S39]; advertising.amazon.com/blog/amazon-ads-mcp-server 404s and the Ads API docs pages returned only a title. Unresolved — A2/A6. - Amazon Ads API marketplace coverage for AE/SA. SP-API coverage is proven [A1-S31]; the Ads API equivalent could not be read (docs pages returned only "Amazon Ads Advanced Tools Center"). Amazon Ads' own Middle East guide discusses Sponsored Products only [A1-S36], which hints at a thinner ad surface — hint, not evidence. - Amazon Seller Assistant. See §3. Capabilities, autonomy tiers, cost and store availability all rest on a competitor's blog [A1-S38]. A2 must verify from Amazon.

Product identity - "Pixie". 006 recorded Pixie as one-click AI Amazon listing design 006:[L87]. gopixie.ai returned an expired certificate and pixie.io returned nothing usable on 2026-09-05. The $9–$69/mo "Pixie" found in a pricing directory is probably a different company. Unresolved; do not cite either way.

Not asked, but load-bearing for the Brief - Nothing found priced or packaged for a pre-revenue seller that does anything other than research. Several products are structurally unusable pre-revenue: Superfuel prices per ASIN off trailing-30-day sales [A1-S6]; AdLabs' free tier needs a live ad account [A1-S16]; 8fig requires $100K+ annual revenue [A1-S28]. This is a real wedge signal, but it is an absence, and absences belong here rather than in §1.

5. What this does NOT cover

• Amazon's and noon's first-party seller AI in any depth. Seller Assistant, Ads Agent, Creative Agent, Campaign Manager beta and Rufus are A2's subject. A1 records only enough to place them in the competitive set, and one unresolved contradiction (§3).
• The technical ceiling. Which SP-API / Ads API / Noon API operations are actually writable, rate limits, ToS on automated console changes — A6. A1 proves only that AE and SA marketplace IDs exist [A1-S31] and that noon's partner APIs exist [A1-S29].
• The regional supply side. MENA SaaS, Noon-native tooling, GCC enablers and Arabic commerce AI — A3. A1 searched globally and found no GCC-built AI seller product; that is a null result from a global sweep, not a regional sweep.
• Courses, communities, agencies, bots. A4. Lumian and Atomic One appear here only because they present as software products.
• The import spine itself. A7. A1 establishes that the software column is empty; it does not map the steps.
• Pricing verification beyond published pages. Every price above is a list price read from a vendor page on 2026-09-05. No discounts, annual commitments, regional pricing, or negotiated enterprise rates were tested.
• Any product that does not present as AI-assisted. Pure repricers, pure OMS, pure freight-forwarders and pure accounting tools were excluded, except Omniful and ChannelEngine, which were kept solely because they are the only software found writing to noon.com.
• Non-marketplace platforms. Salla, Zid, Shopify and own-store tooling are out per SCOPE.
• Walmart, TikTok Shop, Instacart, Target coverage of the platforms above, which several have and which is irrelevant to a Gulf importer.

Triage: what I excluded and why

Per the Scout pattern, exclusions are stated, not silent.