🤖 ChatGPT Wallet gives agents a way to pay—with approval for every purchase


Nvidia changes leadership, llama.cpp formalizes versioning, and Cloudflare opens its agent browser.
The Merpati Post
Daily AI Briefing

Issue · November 14, 2026

ChatGPT Wallet brings approved agent payments into existing checkouts, while Nvidia changes leadership and developer infrastructure becomes more open and predictable.

A hand holds a green approval seal above a terracotta wallet as a payment card passes through an arch; a parcel sits nearby and a small gray pigeon perches at the left.

AI in general

Frontier models, research and policy

3 stories

Score · 99 / 100

ChatGPT Wallet gives agents a way to pay—with approval for every purchase

Source: OpenAI — November 13, 2026

AGENTS · PAYMENTS · COMMERCE

OpenAI is rolling out ChatGPT Wallet through its dots agents, initially for eligible adult ChatGPT Pro subscribers in the US. Users review and approve each purchase with a security check, after which a payment-network token lets the agent complete checkout without seeing the user’s actual card details. This moves agents beyond shopping advice into transactions, but it is not yet unrestricted, autonomous spending.

What matters

  • Before approval, users see the items, total, and selected card; security checks can include Face ID or a fingerprint.
  • Existing card rewards and protections for eligible disputes remain, while merchants retain responsibility for fulfillment, returns, and customer service.
  • Tokenized payments use existing checkout flows, but website blocking and fraud-system declines can still prevent completion.
  • Pre-approved spending within limits is a future direction, not the initial launch behavior.

VerdictREAD FULL — useful for understanding both the authorization model and the changes merchants may need.

Score · 95 / 100

UPDATE — Nvidia confirms Jensen Huang’s resignation and names Colette Kress interim CEO

Source: NVIDIA — November 12, 2026

LEADERSHIP · AI INFRASTRUCTURE · GOVERNANCE

Following our November 9 coverage of Nvidia’s independent investigation, the company has confirmed a consequential change: Jensen Huang has resigned as president, CEO, and board member, effective immediately. CFO Colette Kress becomes interim CEO while retaining her finance role, and the board is searching for a permanent successor. The announcement establishes the leadership transition; it does not resolve the allegations surrounding the investigation.

What matters

  • The permanent CEO search will consider both internal and external candidates.
  • Harvey C. Jones becomes Nvidia’s first-ever board chair.
  • The board says operations, customer commitments, and the product roadmap should continue without interruption; that is management’s stated expectation, not an independently verified outcome.
  • Nvidia’s November 17 earnings call is the next scheduled opportunity to assess operational continuity.

VerdictSKIM — the official statement is short and separates confirmed governance changes from speculation.

Score · 95 / 100

Gemini can analyze Google Wallet spending—but cannot move your money

Source: Google — November 13, 2026

PERSONAL FINANCE · PRIVACY · CONSUMER AI

Google is adding an optional Wallet connection to Gemini for US users, enabling answers informed by spending history. Users can also link bank and credit-card accounts to Wallet through Plaid for broader financial insights. Unlike today’s OpenAI announcement, this is a read-only financial-analysis feature, not a payment authorization system.

What matters

  • Example tasks include finding recurring subscriptions, flagging price increases, and generating spending or budget charts.
  • Google says Gemini cannot move, manage, or invest money through the connection.
  • The rollout covers web and mobile, including Android and iOS; users can disable the connection in settings.

VerdictSKIM — the main decision is whether personalized budgeting help warrants connecting sensitive financial data.

Software engineering

Coding agents, developer tools and infrastructure

3 stories

Score · 99 / 100

llama.cpp reaches v1.0.0 with explicit API and ABI version boundaries

Source: ggml-org / Hugging Face — November 13, 2026

LOCAL AI · DEVELOPER TOOLS · INFERENCE

llama.cpp and its underlying ggml tensor library have released v1.0.0, with a clearer commitment to signaling compatibility changes for downstream developers. Breaking public C API changes will require major-version bumps, while backward-compatible additions and fixes receive smaller increments. The important development is a more predictable integration contract—not a claimed leap in model intelligence or inference speed.

What matters

  • Both projects will make API and ABI incompatibilities visible through major versions.
  • The published rules distinguish breaking public API changes, compatible feature additions, and fixes without API changes.
  • This matters most to teams embedding local inference in applications or maintaining packaging and release pipelines.
  • The announcement describes support for more than 150 model-architecture variants and continued investment in cross-hardware CI.

VerdictREAD FULL — particularly if your product links against these libraries or vendors their headers.

Score · 97 / 100

Cloudflare open-sources Kitesurf, its lightweight browser engine for agents

Source: Cloudflare — November 13, 2026

AGENTS · BROWSER AUTOMATION · OPEN SOURCE

Cloudflare has released Kitesurf under Apache 2.0, allowing developers to inspect, modify, run locally, and deploy its experimental agent-oriented browser engine. It parses HTML, applies CSS, executes JavaScript, and renders screenshots on Workers without a Chromium container. Chrome DevTools Protocol support gives existing clients such as Puppeteer a route into the engine, although compatibility remains a work in progress.

What matters

  • The repository includes the engine, Rust core, DOM shims, renderer, and CLI.
  • Deployment to your own Cloudflare account requires a Workers Paid plan because pages use Dynamic Workers.
  • Cloudflare is targeting lower CPU and memory use than a full browser; the announcement does not establish a universal performance advantage.
  • Test representative sites before treating it as a Chromium replacement—the team explicitly describes heavy ongoing development.

VerdictREAD FULL — worth evaluating if browser startup costs or resource consumption constrain your agents.

Score · 96 / 100

DeepSeek exposes its app’s search, files, safety, and voice capabilities through APIs

Source: DeepSeek — November 13, 2026

API · SEARCH · MULTIMODAL · DEVELOPER TOOLS

DeepSeek is making the capabilities behind its consumer app available to developers, including web search, file input, URL reading, safety checks, and speech services. V4.1-Flash supports built-in features across Messages, Responses, and Chat Completions, with tool calls and context assembly handled server-side. Standalone APIs are also available for teams that want to compose their own workflows.

What matters

  • Search options include cited web/image search in non-thinking mode and search during multi-step reasoning.
  • File inputs can bring PDFs, Office files, and images into conversation context.
  • Tool-use fees are additional to standard token charges; model-token pricing alone will not describe application cost.
  • Public-facing applications must enable safety checks, and developers retain moderation and applicable compliance responsibilities.

VerdictREAD FULL — relevant if you are comparing an integrated assistant backend with assembling separate search, parsing, and speech services.

Design & creative

Creative workflows and user experience

1 story

Only 1 strong recent items found.

Score · 91 / 100

Figma argues for automating creative labor without automating away the thinking

Source: Figma — November 13, 2026

DESIGN · CREATIVE WORKFLOWS · HUMAN JUDGMENT

Figma Weave co-founder Itay Schiff offers a practical distinction for AI-assisted creative work: some friction is tedious production labor, while some is how people develop ideas and judgment. A polished generated result can hide the fact that the creator has skipped the thinking a rough first draft would have forced. This is a new practitioner essay rather than a product launch or a measured productivity study.

What matters

  • Assess each workflow step before automating it: does it merely consume effort, or does doing it help clarify the creative direction?
  • Writing a brief or manually editing a shot can be part of decision-making, not just an execution cost.
  • Generated material still needs selection, shaping, and authorship; surface polish is not evidence that the work is finished.

VerdictREAD FULL — useful for setting team norms around what to delegate and what designers should deliberately retain.

Open-source watch

Projects gaining meaningful traction

3 stories

Project · 01

Codex Security — openai/codex-security

AI · SECURITY · DEVTOOL

OpenAI’s Codex Security project provides a CLI and TypeScript SDK for finding, validating, and fixing security vulnerabilities. It is worth tracking because it aims to cover more of the remediation loop than simply producing a list of suspicious code patterns.

What matters

  • The inventory records 751 stars today on GitHub Trending, a discovery signal rather than evidence of detection quality.
  • The listed npm package is @openai/codex-security.
  • Before production use, evaluate false positives, patch correctness, and permissions on a bounded repository.

VerdictWATCH — relevant security tooling, but trending traction does not establish that automated findings or fixes are trustworthy.

Project · 02

Open Code Review — alibaba/open-code-review

AI · CODE REVIEW · DEVTOOL

Alibaba’s code-review tool combines deterministic pipelines with an LLM agent and produces line-level comments. Its built-in multi-language rules cover issues such as null-pointer errors, thread safety, XSS, and SQL injection, making the hybrid architecture more interesting than a generic “ask an LLM to review this diff” wrapper.

What matters

  • The inventory records 215 stars today on GitHub Trending.
  • The project lists OpenAI- and Anthropic-compatible integrations.
  • Its claims of efficiency and deployment at Alibaba’s scale are project descriptions, not independent evidence of review accuracy.

VerdictTRY — pilot on historical pull requests with known defects to measure useful findings versus review noise.

Project · 03

A2UI — a2ui-project/a2ui

AGENT · DESIGN · UX · DEVTOOL

A2UI is a generative-UI framework in which agents stream layout data and applications render it using native components. It is interesting for products that need agent responses to become structured interfaces rather than long chat messages or arbitrary generated pages.

What matters

  • The inventory lists renderers across React, Angular, Lit, Flutter, SwiftUI, and Jetpack Compose.
  • GitHub Trending recorded 117 stars today.
  • Evaluate component restrictions, accessibility, and validation of agent-provided layouts before exposing consequential actions.

VerdictTRY — a useful prototype path for agent-driven forms, cards, and task interfaces across platforms.

Editor’s note

Fresh, non-duplicative developments with concrete implications for agent permissions, infrastructure stability, developer integration, and the creative judgment worth keeping human.

30-second feedback

How useful was today’s digest?

★★★ Very useful ★★ Somewhat useful ★ Not useful

Your choice opens the short form with your rating filled in. Or share a quick note.


Prefer reading on Telegram? Subscribe to The Merpati Post →

The Merpati Post · Daily AI Briefing

background

Subscribe to The Merpati Post