ChatGPT's new team agents, and what to check first
OpenAI's new team tools can take on recurring work, but access depends on your plan and connected apps need a permissions check. We also look at a cheaper coding model, Meta's small-business agent, and how McDonald's uses AI to recommend prices.
The Big Story
OpenAI News
DevDay 2026 Recap
OpenAI announced more than 20 DevDay 2026 updates, including always-on dots agents, shared ChatGPT Spaces, recurring team tasks, and a Meetings plugin. Dots are available on Pro and Business Premium in eligible markets, while Spaces are on Pro, Business, and Enterprise; team tasks and @ChatGPT in Slack or Microsoft Teams require Business or Enterprise, but teammates using the chat integration don't need individual ChatGPT licenses. Developers also get new Codex workflows, plugin tools, and an Agents API with computer use, though some features are still in preview or limited to higher-tier plans.
If the business already pays for ChatGPT Business, check which team tasks and integrations are available before adding another automation service; review connected-app permissions before enabling agents to act on company data.
For Everyone
TechCrunch AI
OpenAI’s latest features take direct aim at the app store model
OpenAI is adding in-chat app suggestions and interactive extensions to ChatGPT, so users can connect and use third-party tools without leaving the conversation. Its new Sign in with ChatGPT option has 16 launch partners and lets users carry their existing AI allowance into participating apps; an enterprise marketplace has more than 30 partners, including HubSpot and Salesforce. OpenAI also says its Dots agents can use connected apps, though it announced no app-store-style billing system and hasn't said third-party subscriptions will go away.
If customers find your business through an app, watch whether that app becomes available inside ChatGPT, but don't assume a ChatGPT subscription covers its separate fees.
AI TLDR
Eleven v4 — ElevenLabs speech models cover 90+ languages and clone from 10s
ElevenLabs released Eleven v4 and the lower-latency v4 Turbo for text-to-speech, with support for more than 90 languages and voice cloning from 10 seconds of audio. The company says Turbo reaches a median ~150 milliseconds to first speech in its tests (excluding network latency), and that v4 improves emotional delivery, pronunciation controls, and voice consistency across longer recordings. Both models are available in ElevenAgents, ElevenCreative, and the API, with a free account able to try them; the article doesn't give per-use pricing.
Businesses producing multilingual audio or using voice agents can test a few real scripts, especially names and sensitive customer interactions, and compare voice quality, latency, and usage costs before switching.
For Builders
OpenAI News
Introducing GPT-6.1 Sol
OpenAI released GPT-6.1 Sol for coding and multi-step work, pricing its API at $2 per million input tokens, $0.10 per million cached input tokens, and $10 per million output tokens (one-fifth of GPT-6 Astra's standard input and output prices). OpenAI reports stronger results than GPT-6 Sol on its coding, document, and computer-use evaluations, though those tests don't establish how it will perform on any particular business workflow. Sol is available through the API and in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu users, but not yet in ordinary Chat.
Teams paying for automated coding or document workflows can compare Sol's output quality and total cost against their current model on a few real tasks; Chat-only users need to wait for availability there.
TechCrunch AI
OpenAI gives Codex reusable cloud environments that work across devices
OpenAI added reusable cloud development environments to Codex, letting developers resume work across computers, phones and shared team setups with approved permissions. Its updated command-line tool adds voice controls and a view for tracking multiple agents, while the ChatGPT desktop app gets Codex-assisted code review. A separate Codex Security Cloud tool can scan GitHub repositories on demand, on a schedule or after commits and prepare fixes in the cloud.
For a business with an in-house developer, a small repository is a sensible place to test the review and security tools, but keep a human responsible for approving fixes and access.
GitHub Trending
context-mode: Context window optimization for AI coding agents. Sandboxes tool output (98% reduction), persists session memory, and enforces routing across 17 platforms via MCP + hooks.
Context Mode is an open-source add-on for AI coding agents that keeps bulky tool output out of the conversation and saves work history for later retrieval. It runs commands in a sandbox, indexes session events in SQLite, and uses hooks to steer supported agents toward those tools; the project lists integrations across 17 platforms, though hook coverage varies. Its own examples show 315 KB of raw output reduced to 5.4 KB (98% less context used), but those figures are project benchmarks, not independent tests.
If a developer on your team uses an AI coding assistant, ask them to test it on a real project and check whether their chosen assistant supports hooks; there’s no immediate action for businesses that don’t build software.
AI TLDR
Jeeves — PostHog’s 9B decision model reasons before it picks
PostHog published Jeeves, an MIT-licensed, 9-billion-parameter decision model that reasons before returning probabilities for yes/no, multiple-choice, or rating questions through a Jev-compatible API. It scored 0.935 on 231 public JevBench items versus Jev's 0.866, but trailed Jev on general-knowledge tests; the sealed benchmark tier wasn't included. Running it requires Python 3.12 and a CUDA GPU, and PostHog reports a median 3.3 seconds per reasoned request on one H100 (with a 17-second 90th-percentile latency for full reasoning chains).
This is a developer tool, not a ready-made small-business app. A team already building automated routing or triage could test it on its own decisions, but should budget for GPU hosting and check slow-case latency before using it with customers.
For Business
OpenAI News
Introducing dots
OpenAI is rolling out dots, always-on agents with their own cloud computers that can use connected apps, carry context across ChatGPT, Slack, and Teams, and work on tasks while a user is away. The first dot is included on Pro and Business Premium in eligible markets, though deeper work has usage allowances; Enterprise, Edu, and Healthcare users need an admin to enable the beta. Background research is read-only, while actions that change accounts or share information go through review rules and may require approval; OpenAI still advises checking consequential work.
If your plan and market are eligible, test a dot on a low-risk task first, and review its app permissions and approval rules before connecting customer records or letting it send anything.
AI News
How McDonald’s uses AI to recommend menu prices
McDonald’s uses a machine-learning system to recommend menu-item prices for individual restaurants, drawing on transaction data, local competitors and estimates of customers’ price sensitivity. Reuters reports that recommendations go periodically to franchisees, who McDonald’s says retain final control; some franchisees described pressure to follow them, which the company disputes. The system recommends prices by restaurant rather than by individual customer, a distinction from personalized pricing practices the FTC is examining.
If you use pricing software, keep a person accountable for final prices and ask whether it uses customer-level data or competitors’ nonpublic pricing information before adopting its recommendations.
TechCrunch AI
Meta is expanding its AI agent Muse to small businesses
Meta expanded its Muse AI agent to small businesses, with connections to Shopify, QuickBooks, Stripe, Slack, Instagram professional analytics, Facebook pages and Meta ad accounts, among other tools. Meta says Muse can use that context to help with sales, operations and marketing. Muse for Small Business has a free tier with usage limits and a paid subscription for more usage, though the article gives no subscription price.
If you already run sales or ads through Meta, try a limited task on the free tier first and check what customer and financial data each connector can access before linking accounts.
xAI News
Team Bots: AI coworkers that learn from your team
xAI launched Team Bots in public beta for Grok Teams and Enterprise plans, letting a company share an AI bot with common files, instructions, app connections, and learned workflow context. Team members can work with a bot privately or invite it into Slack; xAI describes uses in sales briefings, marketing review, engineering coordination, and read-only data analysis. Insurance company Harper says it built a bot in 24 hours to find lapsed policies and help customers reinstate coverage, saving customers more than $120,000 (a customer-reported example, not a general performance estimate).
If the team already uses Grok's eligible plans, try a bot on a narrow shared workflow, and check which files and credentials it can access before adding it to customer-facing work.