ChatGPT ads arrive for Shopify and HubSpot
Shopify and HubSpot users can now test ads inside ChatGPT. We also look at a practical way to measure AI returns, Claude's unified workspace, and new tools for coding agents and connected homes. Several are early or limited, so the details matter before you commit.
The Big Story
OpenAI News
Reimagining advertising with AI
OpenAI introduced Sponsored Agents, which let users begin a clearly labeled conversation with an advertiser after clicking a ChatGPT ad, with initial testing limited to select U.S. advertisers. New Ads Manager tools can create, update, and analyze campaigns from natural-language prompts, suggest copy and imagery, and optionally adapt or translate ad text for a conversation. HubSpot customers and U.S. Shopify merchants can manage ChatGPT ads in their existing tools now, with Shopify availability expanding to supported international markets September 23.
HubSpot and Shopify users advertising in ChatGPT can test a small campaign now, but should review AI-generated creative, monitor lead quality and conversion cost, and clearly separate sponsored-agent responses from independent assistance.
For Everyone
TechCrunch AI
Anthropic merges Claude chat and Cowork in one interface
Anthropic is combining Claude chat, Cowork, Artifacts, and Claude Design in one interface that automatically routes requests rather than making users select a workspace. New presentation and document tools can create and edit files, export slides as PDF or PowerPoint, share work by link, and let users monitor desktop tasks from mobile. The features will reach Pro and Max plans across web, desktop, and mobile over the coming weeks, with free and team tiers planned later.
Claude Pro and Max users should test the unified workflow on a real document or presentation process, but teams should wait for team-tier availability and confirm sharing controls before adopting it for client work.
For Builders
Hacker News
OpenSpec – A lightweight and configurable AI spec framework
Fission’s OpenSpec is a configurable framework for keeping software requirements, coding agents, and implementations aligned through a specification-driven workflow. Version 1.13.0 supports Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, OpenCode, and more than 33 other tools, with commands to explore a codebase, draft proposals and specs, implement tasks, verify the result, and archive completed work. Fission reports more than 265,000 monthly developers and 68,800 GitHub stars.
Small development teams using AI coding agents can test OpenSpec on one contained feature to see whether explicit proposals and verification reduce rework, but should weigh that benefit against the extra specification maintenance.
TechCrunch AI
Your AI agents can now control your Google Home devices
Google launched early access to a Google Home MCP server that lets compatible AI agents, including Claude and ChatGPT, control connected devices, review camera summaries, inspect event history, and build dashboards through natural-language requests. Setup requires a Google Cloud project, Home MCP configuration, and explicit account permissions. The rollout is limited to U.S. subscribers on the $20-per-month Google Home Premium Advanced plan, with broader availability still unannounced.
Premium Advanced subscribers experimenting with agents can test Home MCP, but they should grant only necessary permissions and treat camera and household event access as sensitive data.
xAI News
Memory in Grok Build
xAI added memory to Grok Build, allowing new coding sessions to reuse project conventions, decisions, and durable facts captured after each completed turn. Notes are stored as Markdown by project, with separate global preferences; Grok excludes secrets, tentative conclusions, current task state, and information already documented in the repository. The new `/memory` command lets users inspect saved notes, while `/dream` periodically organizes them into topic files, and current-session instructions always take priority.
Teams trying Grok Build should inspect its memory regularly and correct stale notes, while continuing to keep authoritative commands and architecture details in repository documentation rather than relying on an assistant’s recollection.
For Business
OpenAI News
How to connect AI usage to business value
OpenAI added ChatGPT Admin Console analytics that connect ChatGPT Work and Codex usage and spending with task categories, model choices, plugin adoption, and engineering outcomes such as Codex-assisted merged commits. Its recommended ROI method starts with a workflow baseline, includes review and correction time, and compares measured gains with AI, setup, training, and support costs. In OpenAI’s hypothetical sales example, 5,520 saved hours produce $207,000 in annual capacity value against $60,000 in first-year costs, or a 245% illustrative ROI.
Before expanding an AI rollout, pick one repeated task, measure its current time and quality, then run a defined trial that counts both subscription costs and the human effort needed to check the output.
TechCrunch AI
Meta expands subscription push with new AI-focused plans
Meta introduced Meta One subscriptions that bundle premium Facebook, Instagram, and WhatsApp features with expanded access to Muse image and video generation and other AI tools. Consumer plans cost $7.99 per month for Core and $19.99 for Premium, though Meta hasn’t published fixed usage limits. Business and creator plans start at $14.99, $49.99, $149, and $499 per month, adding progressively more Meta Business Agent capacity, analytics, scheduling, team access, and account-management features.
Businesses should compare the $14.99 Essential and $49.99 Advanced tiers against their current social-management stack, but get region-specific AI limits and agent-response allowances before committing.
Worth Watching
TechCrunch AI
Anthropic and OpenAI want to embed safety evaluators. Will they really be independent?
Anthropic CEO Dario Amodei proposed embedding independent safety evaluators inside frontier AI labs, and OpenAI CEO Sam Altman said his company would participate. Researchers support deeper access to training checkpoints, logs, and employees because advanced models may recognize and behave differently during safety tests. However, Anthropic and OpenAI haven’t specified who the evaluators will be, how much access and time they’ll receive, or whether contracts will let them publish findings without company control.
Businesses evaluating AI vendors should ask for independently published safety findings and the auditors’ scope, access, and testing window rather than treating an unspecified third-party review as proof of safety.