SumoGrowth AI Insights
Top AI News with Trending AI Tools and B2B Automation Tips
SumoGrowth AI Insights
Meta launches Muse Ember, a low cost model built for the tool calls and routing agents repeat all day
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
Top News: Meta launches Muse Ember, a low cost model built for the tool calls and routing agents repeat all day
Plus: Anthropic releases Claude Fable 5.2 at a 20% lower price as it sets its IPO range
Top AI Tools:
Maxma (https://maxma.ai/) works like an AI teammate in Slack that analyzes paid ad performance, creates ads and landing pages, and launches and optimizes campaigns across Google, Meta, TikTok, LinkedIn, and more, proposing changes for your approval.
Alsona (https://alsona.com/) tracks 32+ buying signals like hiring, funding, tech changes, and ad spend, scores companies against your ICP, and uses AI agents to write personalized LinkedIn and email outreach that books sales calls.
Latch.ai (https://www.latch.ai/) automates click heavy accounting close work without prompts: you do a task on screen and talk it through, and it builds a sidekick that repeats it the same way every time, stopping to ask when something doesn't fit.
Business AI Tip of the Day
Split Every Client Agent Into a Planner and a Low Cost Worker
Welcome back to Sumo Growth AI Insights. I'm Darius, and I'm walking you through this week's AI news the way an operator actually uses it, not as a pile of headlines. Lead Story Meta launches Muse Ember, a low-cost model built for the tool calls and routing agents repeat all day. Meta Superintelligence Lab released Muse Ember on Monday, its smallest and most cost-efficient model yet, designed for what Meta calls the work agents repeat all day. Tool calls, coding subtasks, routing, extraction, and long document parsing. It costs 10 cents per million input tokens and 50 cents per million output tokens, with a 1 million token context window, and Meta claims it delivers more intelligence per dollar than Claude Haiku 5.5 or GPT-6 Luna at any reasoning setting. Ember takes text, images, audio, video, and PDFs, so agents can read screenshots, scan documents, or meeting recordings without a separate model. Meta pitches it as the worker in a two-model setup. A bigger model like Muse Wonder Plans and Ember handles the volume. For agencies, that split is the playbook for client agents that answer leads, tag contacts, and sort tickets all day, where cost per task decides your margin. Plus Story Anthropic releases Claude Fable 5.2 at a 20% lower price as it sets its IPO range. Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on Monday, two versions of the same underlying model with different safeguards. Fable is open to all users, while Mythos is limited to vetted cybersecurity and life sciences organizations. Both generate outputs up to 30% faster than the 5.1 versions and now cost $8 per million input tokens and $40 per million output tokens, down from 10 and 50, with an optional fast mode up to 2.5 times quicker at $16 and $80. The launch lands as Anthropic set an IPO price range of $70 to $78 per share for an implied valuation of about $1.404 trillion at the top end. Anthropic still tells most users to start with Opus 5.5 for everyday coding and knowledge work, so agencies should save Fable for the hardest builds and keep routine client automations on cheaper models. Three more headlines. First, Adobe opens Photoshop, Premiere, and more to outside AI agents through local MCP. At its Max event in Miami, Adobe previewed local MCP integrations that let third-party agents from providers like OpenAI, Google, and Anthropic edit native Photoshop, Illustrator, Premiere, and After Effects files while layers and sequences stay editable. Adobe also expanded its creative agent assistance in beta across its desktop apps, and the MCP integrations enter private beta in early 2027. Second, Fireworks AI says hackers reached customer datasets and model weights. Fireworks AI said attackers who breached a third-party cloud provider reached its development machines on October 24th, exposing billing information and, for some customers, datasets and incomplete model weights. It's the company's second disclosed incident in weeks. Fireworks links it to an LLM jacking group that resells stolen model access, and it says payment card numbers were not affected. Third, Google unveils TPU 8TX for confidential inference and TPU 8i Max for giant context windows. Google announced TPU 8TX, built for secure, confidential inference for government agencies and regulated businesses, and TPU 8i Max, which combines two TPU 8i chips to double compute and memory bandwidth for massive context workloads. That makes four new TPUs this year, and Google has not shared pricing or availability yet. Tools of the Day. Maxma at maxma.ai works like an AI teammate in Slack that analyzes paid ad performance, creates ads and landing pages, and launches and optimizes campaigns across Google, Meta, TikTok, LinkedIn, and more, proposing changes for your approval. Alsona at Alsona.com tracks 32-plus buying signals like hiring, funding, tech changes, and ad spend, scores companies against your ICP, and uses AI agents to write personalized LinkedIn and email outreach that books sales calls. Latch.ai at latch.ai automates click-heavy accounting clotheswork without prompts. You do a task on screen and talk it through, and it builds a sidekick that repeats it the same way every time, stopping to ask when something doesn't fit. Business AI tip of the day split every client agent into a planner and a low-cost worker. Meta built Muse Ember for one job, handling the repeated steps inside an agent while a bigger model does the planning. Most client agents run every tag, lookup, and summary on a frontier model, even though only a few steps actually need deep reasoning. So split one client agent into two tiers this week. Map every step the agent runs on a typical lead or ticket, from reading the message to updating the contact. Mark the judgment calls, like qualifying a lead, calming an upset customer, or writing a custom quote for your bigger planner model. Send repeated bounded steps like tagging, field extraction, routing, and call summaries to a low-cost worker model. Have the planner write clear rules for each worker step. Then let the worker apply them to every contact. And track cost per conversation for a week before and after so you can show the client the savings. Cheaper steps mean you can run more work per lead without raising your retainer, which is an easy edge when clients compare quotes. Pro tip When a worker's step hits something unclear, like a message that fits no tag, route it back to the planner instead of letting the worker guess. That's your sumo growth AI Insights briefing. Split one client agent into two tiers this week, and I'll catch you on the next issue.