It's May 18. Q4 holiday briefs start in 6 weeks.
The brands writing 2026 plans right now are the ones who got mobile wrong in 2018: optimising for the channel they already understand.
Agent traffic is the new mobile. You don't catch up in November.
In March, OpenAI quietly moved Instant Checkout off ChatGPT and back to your site.
That's not a downgrade. That's the AEO playbook flipping.
If your PDP can't be parsed by an agent in under 2 seconds, you're invisible now. Half the brands I audit still load 4MB of hero video before the price renders.
@OpenAI@ChatGPTapp
@EBogomolovs Thats the right frame. The trap is letting the three define the spec. You end up with OAuth scopes instead of an open wire format. Keeping ACP brand-led is the whole game.
The reason I shipped @cresva/acp-mcp open source: the agent commerce protocol won't matter if 3 closed platforms control the integration layer.
Brands need an open standard their stack can implement once and have it work everywhere. That's the bet.
A brand asked me last week: "why would you give away the integration layer for free?"
Because the integration layer isn't where the value is. The agent stack on top of it is.
Closed protocols protect weak products. Open ones force you to build something actually worth paying for.
Three protocols competing for agent commerce right now: ACP (OpenAI/Stripe), AP2 (Google), UCP (Visa).
Two are controlled by ad platforms. One is controlled by a payment network. None of them are controlled by the brands that have to integrate with all three.
That asymmetry is why open implementations matter.
Every protocol war in commerce history ended the same way: the closed version won short-term, the open version won the decade.
EDI. OAuth. REST. Even payment rails.
Betting ACP follows the same path. That's why @cresva/acp-mcp is MIT licensed and not "available on request."
Ran this test 4 times today. Same prompt to ChatGPT: "best running shoes under $150."
4 fresh sessions, no memory. Got 4 different shortlists.
Nike showed up twice. A brand I'd never heard of, twice. Asics, zero times.
The "ranking algorithm" everyone is optimizing for doesn't exist yet. We're all guessing in public.
@vikrambuilds Realizing the security fix you're confidently shipping is based on a runbook claim YOU wrote 6 months ago and was just wrong.
Solo founder edition.
@FlippedRay Building the product to a point where the marketing writes itself. Solo founder with 100% of time on the product is the cheapest moat to build before the company has revenue.
Alone. Tried with a cofounder, didn't work out. Back to solo.
Solo is harder but cleaner. No alignment risk, no equity dilution, no "are we still building the same thing" conversations.
The tradeoff is real: nobody to disagree with you when you're wrong. You learn to be your own critic.
@1Umairshaikh Building something nobody wants kills 90%.
No distribution kills the other 10% who built the right thing but couldn't get it in front of people.
@TTrimoreau Distribution. Twice over.
Best product with no distribution = 0 customers.
Worst product with great distribution = funded and acquired.
Build the distribution muscle while the product is mediocre. By the time the product is great, the distribution channel is your moat.
@KaiXCreator The honest answer most founders won't give: not much.
Claude can code, plan, write, decide. Better than we can. What we bring is taste, commitment, and the legal liability. That's still worth something.
But pretending I'm "the visionary" is just ego.