Siml says agents run 1,000 storefronts. The number to pressure-test is Adobe's 66%

Over 1,000 is a countable claim, which is the reason to start there rather than with the dropout narrative and the investor list ^[1][3]. The tests are dull ones. How many of those storefronts took an order last week? What share of daily actions execute without a human approving them, and what does the vendor's own logging count as an exception? Since the agents are described as holding the state of the whole business and acting on it continuously ^[13], the metric that matters is not stores under management but interventions per store per week. The account carries none of it: no revenue per store, no category mix, no platform breakdown, no autonomy share ^[14].

Customs is where being wrong gets expensive. With the de minimis exemption suspended in 2025 and confirmed in February 2026 with no restoration date, roughly four million packages a day need formal treatment, each requiring a classification code, a country of origin, a material description, and a landed cost recomputed per shipment ^[5]. An agent doing that work is not drafting product copy; it is generating regulated filings at volume.

The demand-side evidence is the strongest part of the case, and it is not Siml's. Adobe Analytics put traffic from AI sources to US retail sites up 393% year over year in the first quarter of 2026 ^[6]. In March 2025, those visitors converted 38% worse than paid search and email; by March 2026 they converted 42% better, with revenue per visit 37% higher [c7a][c7b]. That is an eighty point turn in relative conversion inside twelve months ^[8], and it means the buying side re-plumbed itself faster than the selling side noticed.

Which makes the 66% sit oddly beside the growth. Adobe scored retail product pages at 66% machine readability, so about a third of the content on the page where the purchase decision happens is unreadable to the models sending that traffic ^[9]. Agents converting well against pages they can only partly parse suggests the traffic arrives pre-qualified, not that the pages are doing the work.

The forecasts deserve less weight than they will get. Worldwide retail ecommerce is projected at $6.9 trillion in 2026 ^[10], and McKinsey puts agentic commerce at $3 to $5 trillion globally by 2030 ^[11]. Measured against that 2026 base, the projection covers 43% to 72% of all online retail ^[12]. Different years, different definitions of what counts as agent-driven, and a range wide enough to accommodate any outcome, which is what makes it easy to quote in a deck.

Zuhayr Abdullazhanov's own summary is the least promotional sentence on offer: "The tools got prettier. The work did not get easier." ^[15] He and Nurtilek Raimzhanov were fifteen when they started selling hand sanitizer online ^[2], and the pair published work on agent orchestration before building this ^[4]. The honest read of their own market data is that the readability figure is the part an operator can move without buying anything from them.