Get tomorrow's brief in your inbox
Today: Hootsuite shipped Wisdom, an AI agent that monitors 150M social data sources to help small businesses anticipate trends instead of chasing them. Meta quietly launched Pocket, an AI game builder that turns written prompts into playable apps. Anthropic is exploring custom chips with Samsung to reduce dependence on Nvidia.
Hootsuite just shipped Wisdom, an AI agent that gives small businesses something they've never had before: the ability to know what's trending on social media before their competitors do. Built on 15 years of proprietary social data and 150+ million monitored sources, Wisdom synthesizes live social signals and delivers actionable recommendations in real time.
For small business owners already stretched thin, this matters. Instead of manually monitoring social feeds or waiting for weekly reports, you can ask Wisdom questions like "what are people saying about our competitor's new product?" or "what content ideas are trending in our industry?" and get immediate, context-aware answers. The system can identify emerging trends from smaller conversations, assess risks, and generate content ideas without you having to sift through mountains of data.
Wisdom is part of a broader rollout Hootsuite calls Social OS, which consolidates publishing (Perch), customer care (Nest), and employee advocacy into one platform. For teams managing multiple social functions, this means significant time savings. The integration with Model Context Protocol (MCP) lets you pipe Hootsuite's social insights into other AI tools you already use, so the data doesn't sit isolated.
The catch: adopting a new system takes time and training, especially for resource-strapped teams. You'll need to invest upfront to learn how to use these tools effectively. Data privacy is another consideration, particularly if you're handling customer information. But if you're currently drowning in social media noise or struggling to stay ahead of trends, Wisdom gives you an intelligence advantage your competitors may not have yet.
Meta Pocket: AI Game Builder Goes Live Meta quietly launched Pocket, an app that lets you create small interactive games and apps using text prompts. Type "make me a trivia game about space exploration" and Pocket generates a playable "gizmo" you can share. The app includes a discovery feed where you can play what others have built. It's free, based on Meta's acquisition of vibe-coding platform Gizmo earlier this year. Still in experimental phase (Meta hasn't officially announced it), but if you want to prototype interactive content without coding, worth checking out. Available on App Store and Google Play.
Simon Willison's llm-coding-agent: Build Your Own Claude Code
Developer tool builder Simon Willison shipped a new coding agent built on his LLM library. It's a Python-based agent with tools for reading/editing files and executing commands. You can run it with uvx --prerelease=allow --with llm-coding-agent llm code. Supports approval workflows (ask before running commands) or YOLO mode (run everything). Built using test-driven development with Claude's help. If you're a developer who wants to understand how coding agents work under the hood or customize one for your workflow, this is a solid starting point. Details here.
Springboards' Flint LLM: Escaping AI Groupthink Australian startup Springboards built an LLM called Flint specifically trained to avoid the predictable, groupthink responses you get from ChatGPT, Claude, and Gemini. Ask mainstream models for "a random number between 1 and 10" and you'll usually get 7. Flint generates more varied responses to open-ended questions like "where should I go in Europe?" If you're using AI for brainstorming, creative work, or planning (not coding or research), this diversity matters. Pricing and access details not yet public. MIT Tech Review coverage.
Microsoft Launches $2.5B AI Deployment Company Microsoft announced Microsoft Frontier Company, a new division backed by $2.5 billion and 6,000 engineers focused entirely on deploying AI for enterprise clients. This is forward-deployed engineering at scale: experts who embed with your company to make AI projects actually work, not just run proof-of-concepts. Early clients include London Stock Exchange, Unilever, and Land O'Lakes. If you're mid-market or larger and struggling to move AI pilots into production, this could be the implementation support you need. Amazon announced a similar $1B program two days earlier. The FDE model is becoming standard across the industry. Read more.
Woodside Energy: How to Deploy AI in High-Stakes Industrial Environments Energy company Woodside Energy has been using AI for predictive analytics, optimization, and maintenance across its plants for years. Their "Startup Advisor" AI copilot helps operators manage the complex process of starting LNG plants. Key lesson: they didn't bolt AI onto existing processes. They redesigned workflows around how AI augments human expertise. "Think big, prototype small, scale fast" is their approach. If you run operations where mistakes are expensive (manufacturing, logistics, construction), this case study shows how to deploy AI without replacing skilled workers. Full interview at MIT Tech Review.
OpenAI Proposes 5% Equity Donation to US Sovereign Wealth Fund OpenAI CEO Sam Altman proposed giving 5% of the company's equity to a US sovereign wealth fund, with other AI companies (reportedly Anthropic, Google, Meta) doing the same. The idea: public ownership of AI, with returns distributed to citizens. Still preliminary, would require congressional approval. Sen. Bernie Sanders proposed a more aggressive version in June: a 50% tax on AI company stock deposited into a public fund. If you're tracking who controls AI development and how that affects your access to tools, this is worth watching. A public fund could mean broader access or more regulation depending on how it's structured. Financial Times report.
Meta CEO: AI Agents Not Progressing as Fast as Expected Mark Zuckerberg told staff at an internal town hall that AI agent development hasn't "accelerated in the way" executives expected. Meta laid off 8,000 employees earlier this year and reassigned 7,000 to AI groups, betting heavily on AI replacing human work. Zuckerberg said the "upside hasn't come to fruition yet" but expects improvements in 3-6 months. Meta is spending up to $145B on AI infrastructure this year. Translation: automating complex work with AI is harder than the hype suggests. If you're planning workforce changes around AI capabilities, temper expectations and keep humans in the loop longer than you think. Reuters coverage.
Anthropic Exploring Custom Chip Partnership with Samsung Anthropic is in talks with Samsung about developing custom AI chips, following similar moves by OpenAI (Broadcom partnership for "Jalapeño" chips), Amazon, and Google. The goal: reduce dependence on Nvidia and optimize hardware for specific tasks. Anthropic hasn't decided what the chip will do or how powerful it will be. For SMBs, this is infrastructure news, but it signals a long-term trend toward more chip diversity and potentially lower inference costs as competition increases. The Information report.
Meta Exploring Cloud Infrastructure Business Meta is exploring two ways to monetize its massive AI compute infrastructure: selling access to models hosted on Meta's servers or selling raw computing power (similar to CoreWeave and Nebius). Not trying to compete with AWS/Azure/GCP directly, more likely targeting AI-specific workloads. Stock jumped 9% on the news. For businesses shopping for AI compute, this could mean another vendor option and potential price competition. Watch for announcements in Q4. Bloomberg coverage.
Use AI to Plan Dates (or Client Meetings) Across Unfamiliar Cities PR founder Jeff Weisbein uses OpenClaw (Claude's coding agent) to research restaurants and activities across South Florida neighborhoods he doesn't know. Instead of Googling "best happy hour Fort Lauderdale" and clicking through 10 listicles, he asks the agent to compile options with links and explanations customized to the type of meeting or date. You can do the same for client dinners, team offsites, or conference travel. Prompt: "Find 3 restaurant options in [neighborhood] for a casual business dinner, max $40/person, good for conversation, include parking info and why each is a fit."
DSPy for Prompt Optimization Developer Simon Willison used DSPy (a framework for optimizing LLM prompts) to improve the system prompt for Datasette Agent. DSPy evaluates multiple prompt variations against test cases and identifies which performs best. Key insight: including column names in schema listings reduced error-retry loops. If you're building AI features and your prompts feel brittle, DSPy can systematically improve them. Requires Python coding skills but the results are measurable. Details here.
Three patterns across today's stories: First, major AI companies (Microsoft, Amazon, OpenAI, Anthropic) are shifting from selling tools to selling implementation services. The bottleneck isn't technology anymore, it's deployment expertise. Second, the predictability problem in LLMs is getting attention. When every chatbot gives the same answer, differentiation comes from models trained for variety (like Flint) or heavily customized prompts. Third, chip diversification is accelerating. Anthropic/Samsung, OpenAI/Broadcom, and potential Meta compute offerings all signal a move away from Nvidia dependence. For SMBs, this means more vendor options and eventual price competition, but not for another 12-18 months.
Watch for Microsoft Frontier Company client announcements this month. If they land recognizable mid-market brands (not just Fortune 500), that validates the FDE model for companies your size.