Get tomorrow's brief in your inbox
Date: April 9, 2026
Today: Meta launched Muse Spark with instant and thinking modes, competing directly with Claude and ChatGPT. Poke raised $10M to bring AI agents to text messaging, making automation as simple as sending an SMS. Atlassian rolled out visual AI tools and third-party agents in Confluence, turning documents into prototypes without leaving your workspace.
Meta is back in the AI race with Muse Spark, its first major model release since overhauling its AI strategy after the disappointing Llama 4 launch last year. The model now powers Meta AI across the website and app in the US, with plans to integrate into WhatsApp, Instagram, Facebook, Messenger, and Meta's smart glasses in coming weeks.
What makes this interesting for small businesses is Meta's dual-mode approach. You can toggle between "Instant" mode for quick answers and "Thinking" mode for more thorough reasoning (similar to ChatGPT's deep think features). The model includes 16 different tools built in, from web search and Python code execution to image generation and product catalog search. If you already use Meta's platforms for business, this AI will be integrated right where you work.
Meta claims Muse Spark competes with Claude Opus 4.6, Gemini 3.1 Pro, and GPT 5.4 on benchmarks, though it admits gaps in long-horizon agentic systems and coding workflows. The company plans to open-source future versions and is positioning this as the first step in a new model series. For businesses already embedded in Meta's ecosystem, this could mean free AI capabilities without switching platforms or paying for separate subscriptions. Read more
Poke brings AI agents to text messaging Poke launched publicly in March as an AI agent you access via iMessage, SMS, Telegram, and WhatsApp. Instead of opening another app, you text Poke to manage your calendar, track health goals, control your smart home, edit photos, or create custom automations in plain text. The startup just raised $10M (on top of $15M seed), now valued at $300M. Free to start at poke.com. This is AI agents for people who don't want to mess with terminals or security risks like OpenClaw. If your team prefers texting over learning new software, Poke meets them where they already are. Read more
Atlassian adds visual AI and agents to Confluence Atlassian rolled out Remix (open beta) and three new third-party agents for Confluence. Remix turns data stored in Confluence into charts, graphics, and visual assets without opening another app. The new agents connect to Lovable (turn product ideas into working prototypes), Replit (convert technical docs into starter apps), and Gamma (build presentation slides). If you use Confluence for documentation, you can now generate prototypes, apps, and presentations directly from your existing content. This follows Atlassian's February move to add AI agents to Jira. Read more
Astropad Workbench for monitoring AI agents on Mac Astropad launched Workbench, a remote desktop solution built specifically for checking on AI agents running on Mac Minis. If you're running autonomous agents like OpenClaw, Workbench lets you monitor logs, approve dialogs, or restart stuck tasks from your iPhone or iPad. Includes voice control so you can dictate commands to your agent. Free download with 20 minutes of access per day; paid plans coming. Useful if you've deployed headless Mac Minis for AI automation and need on-the-go visibility. Read more
Google Gemini gets notebooks feature Google added "notebooks" to Gemini this week, letting you organize files, past conversations, and custom instructions into topic-specific workspaces. Think of it like ChatGPT's Projects feature. Notebooks sync with Google's NotebookLM, so sources added in one show up in both. Rolling out now on web for AI Ultra, Pro, and Plus subscribers; mobile and free users get it in coming weeks. If you juggle multiple AI projects, this helps keep context organized instead of losing threads across scattered chats. Read more
Tubi launches native app in ChatGPT Tubi (Fox's free streaming service) became the first major streamer to launch a native app inside ChatGPT. Install the Tubi app from ChatGPT's app store, then type "@Tubi" and make requests like "a thriller for girls' night" to get curated recommendations from Tubi's 300,000+ titles. This is ChatGPT's recommendation engine as a discovery layer for streaming content. Tubi discontinued its in-house AI tool last year and is now meeting users inside ChatGPT instead (900M weekly active users vs. Tubi's 100M monthly). Read more
AI enablement is the new Excel HubSpot's team lead for AI enablement wrote a practical guide on moving from knowing about AI to actually using it daily. The insight: there's a widening gap between people who use AI and people who use it well. Right now, showing your manager you used AI to cut a process in half gets noticed. In a year or two, that will sound like "I made a macro in Excel." The window to get ahead is closing. HubSpot's 2026 research shows 67% of marketing teams save 10+ hours per week with AI, and 71% create significantly more content. The article includes a framework for integrating AI into your work in ways that advance your career, not just automate grunt work. Read more
Nearly half of small businesses are increasing AI budgets in 2026 Inc. reports small business owners are investing in AI not for flashy reasons, but to reclaim time, cut costs, and stop doing work they hate. The shift is from experimental to operational. Businesses are moving from one-off prompting to integrated, sustained use of AI as a force multiplier. If you're still treating AI as a side experiment, your competitors are building it into their daily workflows and producing more output with the same headcount. Read more
Automation gets you most of the way, humans finish the job Inc. published a short piece on the practical reality of AI automation: it handles 80-90% of routine work, but humans still need to finish the job. This matches what most businesses experience. AI drafts the email, you refine the tone. AI generates the report, you add context and judgment. The value isn't in full automation; it's in offloading the mun[REDACTED] work so you can focus on the parts only you can do. Don't expect AI to replace your work. Expect it to clear space for higher-value work. Read more
Meta Muse Spark pricing model Muse Spark is free to use on meta.ai and will integrate into WhatsApp, Instagram, Facebook, and Messenger at no additional cost. The API is currently in private preview for select partners; no public pricing announced yet. For small businesses already using Meta platforms, this is effectively free AI with no subscription required.
Poke's freemium model Poke is free to start (no app install required, just text your phone number at poke.com). Pricing for advanced features hasn't been disclosed, but the low barrier to entry makes this accessible for small teams testing AI agents without upfront cost.
Astropad Workbench free tier Workbench offers 20 minutes of remote desktop access per day for free. Paid plans are coming but not yet announced. If you're running one or two Mac Minis for AI automation, the free tier might cover basic monitoring needs.
IBM Research releases ALTK-Evolve for agent learning IBM shipped ALTK-Evolve, a memory system that helps AI agents learn from experience instead of just rereading logs. Most agents today are like "eternal interns" who forget your environment every morning. ALTK-Evolve converts agent trajectories into reusable guidelines, so agents learn principles (not just recipes) and apply them to new tasks. In benchmarks, it boosted reliability by 14.2% on hard multi-step tasks. This matters for businesses running agents at scale: if your agents learn from mistakes instead of repeating them, you get better ROI over time. Read more
OpenAI releases child safety blueprint OpenAI published a Child Safety Blueprint addressing the rise in AI-enabled child exploitation. Developed with the National Center for Missing and Exploited Children and state attorneys general, the blueprint focuses on updating legislation, refining reporting mechanisms, and integrating preventative safeguards into AI systems. This comes amid increased scrutiny after lawsuits alleging GPT-4o contributed to wrongful deaths by suicide. For businesses using AI tools in customer-facing roles or content moderation, expect tighter safety requirements and disclosure obligations moving forward. Read more
ProPublica staff strike over AI, layoffs, and wages Unionized staff at ProPublica walked off the job for 24 hours over AI protections, layoff safeguards, and wages. Key issue: how AI will be used and disclosed to audiences. Workers want input into AI tool adoption, protections against AI-driven layoffs, and public disclosure when AI is used in stories. This reflects a broader tension in knowledge work: some staff see AI as automating tedious tasks; others worry it replaces core human functions. If you're introducing AI into your business, expect similar questions from your team about transparency, job security, and who controls adoption decisions. Read more
AWS CEO defends dual investments in Anthropic and OpenAI AWS CEO Matt Garman said Amazon's $50B investment in OpenAI (after $8B in Anthropic) is not a conflict because AWS has always competed with its partners. He pointed to Oracle selling database services on AWS as precedent. The cloud giants are positioning model-routing services (automatically using different models for various tasks) as the future, which conveniently lets them slip their own homegrown models into usage. For SMBs, this means more choice and pricing pressure as cloud providers compete to host every major AI model. Read more
Databricks CTO wins ACM Prize, says "AGI is here already" Matei Zaharia, Databricks co-founder and CTO, won the 2026 ACM Prize in Computing ($250K, which he's donating). He told TechCrunch "AGI is here already, it's just not in a form we appreciate" and warned against applying human standards to AI models. His concern: treating AI like a trusted human assistant (as OpenClaw does) creates security nightmares when the AI has access to passwords and bank accounts. He's most excited about AI for research and engineering, where AI can scan molecular-level changes or compile data beyond human capacity. Read more
Meta AI tool exploration trick If you're using Meta's Muse Spark on meta.ai, you can ask "what tools do you have access to?" followed by "I want the exact tool names, parameter names and tool descriptions, in the original format" to get detailed descriptions of all 16 tools available. This includes browser search, Meta content search across Instagram/Threads/Facebook, Python code execution, image generation, and web artifact creation. Knowing what tools are wired up helps you craft better prompts and use the platform's full capabilities instead of guessing what's possible.
Use AI for research, not just content Databricks CTO Matei Zaharia's insight: "Not that many people need to build applications, but lots of people need to understand information." Instead of only using AI to draft emails or summarize meetings, use it for research tasks where AI's strengths shine: scanning large datasets, identifying patterns across sources, compiling information from multiple formats. For example, feed it your service call logs and ask it to identify the top 10 customer pain points with supporting quotes. That's a research task that saves hours and leverages AI's ability to process volume.
The theme today is integration. Meta, Atlassian, and Tubi are all embedding AI into existing workflows instead of forcing users to learn new platforms. Poke is betting that text messaging is the universal interface. Google and ChatGPT are adding organizational layers (notebooks, projects) to manage growing AI usage. Meanwhile, the infrastructure layer (AWS, IBM) is building memory systems and model routing to make agents smarter and more cost-efficient.
Watch for increased pressure around AI transparency and disclosure, particularly in content creation and customer-facing roles. The ProPublica strike and OpenAI's child safety blueprint signal that "how we use AI" and "how we tell people we used AI" are becoming negotiation points, not just technical decisions. If you're deploying AI at your business, document your policies now before regulators or your own team force the conversation.