OpenAI and Google launch rival AI checkout protocols ACP and AP2 for online stores

·1 views

OpenAI has released ACP and Google has released AP2, two competing protocols that allow AI agents to complete purchases on behalf of users at online stores. ACP works by creating a managed checkout session via API calls, where the agent interacts with line items, shipping, and totals while payment is processed through a Shared Payment Token charged via Stripe. AP2 takes a different approach, requiring merchants to sign a short-lived JWT containing a W3C PaymentRequest cart, with settlement handled through the buyer's credentials provider and payment network. A key implementation challenge with AP2 is that the cart hash must be byte-identical across all systems, requiring canonical JSON formatting per RFC 8785 rather than standard serialization methods. Merchants looking to support AI-driven checkout now face a choice between the two protocols, each tied to different distribution ecosystems — ChatGPT for ACP and Google for AP2.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Developer compares Nuxt and SvelteKit performance using identical task apps

A developer built the same task management app twice — once in Nuxt 4.5 with a version 5 compatibility flag and once in SvelteKit using experimental remote functions — to compare their network behavior. The key difference found was that adding a task in Nuxt triggered two separate HTTP requests totalling around 1,100 ms, while SvelteKit handled the same operation in a single response of roughly the same duration. Plain page refreshes were nearly identical between the two frameworks, at approximately 447–448 ms each. The author noted that SvelteKit's remote functions consolidate the mutation and data-read into one round trip, whereas Nuxt follows a more explicit invalidate-and-refetch pattern. Since both Nuxt 5 and SvelteKit's remote functions remain unreleased or experimental, the comparison reflects future directions rather than stable, production-ready features.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

What Developers Face When Learning Go and JavaScript Simultaneously

Developers working with Go and JavaScript at the same time often encounter syntax muscle-memory conflicts, such as accidentally writing := in a JS file or const in a Go file. Beyond syntax, the two languages differ fundamentally in type handling: JavaScript allows implicit type coercion, while Go's compiler rejects type mismatches before the code even runs. Go also assigns zero values to all uninitialized variables, unlike JavaScript's undefined or null, which can confuse developers accustomed to JS-style truthiness checks. Error handling diverges sharply as well, with Go relying on explicit err != nil checks after function calls rather than try/catch or Promises. Perhaps the deepest adjustment involves concurrency, since JavaScript uses a single-threaded event loop with async/await, while Go supports true parallel execution through goroutines and channels.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Site-list tool gains instant feedback, error highlights, and auto-deselect in v1.6.8

Version 1.6.8 of a site-management tool introduces three UX improvements to its bulk maintenance workflow, which typically handles 10–20 sites over several tens of minutes. Previously, thumbnails and plugin badges only refreshed after an entire bulk run finished; now each site updates instantly upon its own completion, without affecting sites still in progress. A new error-highlighting feature applies a red border and background to any site that returns a failure marker, persisting for 24 hours and resetting at the start of a new maintenance session. Error detection deliberately uses a single unambiguous log marker to avoid false positives, accepting that a missing marker leaves a site green rather than risk incorrectly flagging a successful one. Finally, selected checkboxes are now automatically cleared after a run completes, preventing users from accidentally re-targeting the same sites in a subsequent operation.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How to Evaluate AI Agents in Production Before You Have Any Labeled Data

Developers deploying AI agents often face a 'cold-start problem': no labeled datasets or historical traces exist to benchmark agent performance from day one. A framework proposed by a developer on DEV Community argues that meaningful evaluation gates can be built without any labels by focusing on independent, unforgeable signals rather than costly model-as-judge scoring. The approach organizes checks into three tiers, prioritizing externally verifiable facts — such as whether a file exists, code compiles, or output is non-empty — over statistical baselines and AI-generated scores. Tier 1 and Tier 2 checks require zero labeled examples because they test whether an output is real, not whether it is good. The author warns that skipping these structural checks in favor of LLM-based scoring introduces circularity and leaves agents running ungated in production until failures become visible.

0 comments Read more at DEV Community