1. Highlights of the week
Between Oct 17 and Oct 23, 2026, agents and AI apps moved through infrastructure, controls and pricing more than through new flagship products. Five conclusions follow; each points to the chapter that holds the evidence, with status judged against Oct 24, 2026.
1. In office, search and creative, one first-party launch was dated, and most of it is still a roadmap. Zoom announced its Unified Context Layer on Oct 20, but only ZoomMate Agent Teams is "available now"; six further features are due from October 2026 to Q1 2027. The rest is a ranked listing with no stated prices: Product Hunt's open weekly board put Beam.win first with 250 votes, and Kai Resolve by Gleap, with 119 votes, was Launch of the Day on Oct 19. [Chapter 2]
2. In coding, the new things were plumbing, and the money moved from plans to meters. Adyen released an open-source Agent Marketplace on Oct 22 that compiles one definition for eight tools, and Daytona announced agent sandboxes on Oct 20. Cursor's X Subscription has run from $8 to $200 a month since Oct 12, with Claude billed outside the plans, while OpenAI's supply of models to Cursor is reported to end on Nov 12. [Chapter 3]
3. Established assistants shipped controls and migrations, not new models. OpenAI began rolling out ChatGPT Locked Chats on Oct 20, which adds a PIN or biometric lock but not end-to-end encryption. Anthropic's Claude API version 2026-10-21 is opt-in, and on it sampling parameters return a 400 error. Meta paused Muse chat sharing after 6,800 links were found exposed. [Chapter 4]
4. The growth signals are strongest for Meta Muse, and they are estimates. Sensor Tower, relayed by 9to5Mac on Oct 20, puts Muse at 10M+ US downloads in 40 days, with day-7 retention of 30% against 19% for peers. Company ARR statements are rarer: Podium said it passed $150M in AI ARR on Oct 22. A RevenueCat benchmark has AI apps keeping 21.1% of annual subscribers against 30.7% for non-AI apps. [Chapter 5]
5. Funding was concentrated, acquisitions were thinly disclosed, and agent pricing is settling on seat plus meter. Daytona raised a $60M Series B at $1B on Oct 20, while Kapital's $30M deal for Woz was the only acquisition with a stated price. Per-resolution list prices run from $0.50 to $2.00 as of Oct 18, and Oracle meters its new Agent Studio in "AI units" without publishing a rate. [Chapter 6]
1.1 Products at a glance
| Product | Scenario | What it does | Price | Status (Oct 24) | Source |
|---|---|---|---|---|---|
| ZoomMate Agent Teams | Office | Specialized agents that work together on shared meeting and document context | Not stated | Available from Oct 20, 2026 | Zoom |
| Kai Resolve by Gleap | Support | Support agent that "does the digging" | Not stated | Launched; Launch of the Day, Oct 19 | Product Hunt |
| Adyen Agent Marketplace | Coding | Open-source CLI that syncs skills, hooks and MCP servers to 8 tools | Free (Apache 2.0) | Released Oct 22, 2026 | Adyen |
| Daytona agent sandboxes | Developer infrastructure | Sandbox computers where agents run code | Not stated | Shipping; announced Oct 20, 2026 | Daytona |
| Cursor X Subscription | Coding | Cursor and Grok in one plan; Claude billed separately | $8 to $200 a month | In effect since Oct 12 (context) | Fello AI |
| ChatGPT Locked Chats | Assistant | PIN or biometric lock with separate memory | Included in Free to Edu plans | Rolling out from Oct 20, 2026 | OpenAI via TechCrunch |
| Claude API 2026-10-21 | Developer platform | New API version with breaking changes; new SDK majors | Not stated | Released Oct 21, opt-in | CellCog |
| Meta Muse | Personal agent | Runs errands from a chat thread on its own cloud computer | Free for most uses; US only | Shipping since Sep 8 (context); sharing paused | Fox 10; Dataconomy |
2. New products: office, search and creative
Chapter 1 gives the week's headline conclusions; this chapter looks at the new office, search and creative products behind them. A product counts as a window product only if it launched or was materially updated between Oct 17 and Oct 23, 2026. Status is judged against Oct 24, 2026.
| Item | Key figure | Date | Status | Source |
|---|---|---|---|---|
| Zoom Unified Context Layer, ZoomMate Agent Teams | Agent Teams "available now"; 6 more features due Oct 2026 to Q1 2027 | Oct 20, 2026 | Announced; Agent Teams available | Zoom |
| Beam.win (agents across Macs) | 250 votes, rank 1 | Week of Oct 19 | Launched; price not stated | Product Hunt |
| readywhen (WhatsApp assistant) | 156 votes, rank 2 | Week of Oct 19 | Launched; price not stated | Product Hunt |
| Kai Resolve by Gleap (support agent) | 119 votes; Launch of the Day | Oct 19, 2026 | Launched | Product Hunt |
| Wally by Willow (writing agent) | 104 votes, rank 7 | Week of Oct 19 | Launched; price not stated | Product Hunt |
| Notion Auto (rebuilt Notion agents) | Runs "for hours at significantly lower cost" (unquantified) | Oct 22, 2026 | Listed as launched | StartupCorners |
| Yelp Assistant, Fall 2026 release | Booking at 500,000+ restaurants | Oct 20, 2026 | Announced; iOS | Yelp (via PR commentary) |
| Kiwi.com AI Search | UK/EU available | Page updated Oct 22 (launch date not stated) | Beta | Kiwi.com Magazine |
| Creative Claw (ChatGPT app) | $10 = 1,000 credits | Quoted Sep 27, 2026 | Context, pay as you go | Creative Claw |
2.1 Zoom: the one primary-source office launch
Zoom Communications (NASDAQ: ZM) announced a Unified Context Layer at Zoomtopia on Oct 20, 2026. It links meeting history with documents and data from Zoom and connected third-party apps, so people and AI agents see the same context. Only one piece is live. ZoomMate Agent Teams, specialized agents that work together, is "available now" (company statement, Oct 20). The rest is a schedule:
- My Notes Workflows (meeting summaries into Zoom Chat and Salesforce tasks): expected October 2026, so still pending on Oct 24.
- Enriched notes and the DTEN My Notes device: November 2026.
- Follow the meeting and the new Zoom Chat: December 2026.
- Bota My Notes device: Q1 2027.
That is 6 features still to come, 5 of them due after Oct 31. The release text in the material states no price, plan tier or region.
Zoom also gave figures with unstated bases: Virtual Agent customer count up 256%, and 72% of Zoom AI users rating business impact high or extremely high against a 63% market average. The footnoted survey and baseline are not in the material, so treat them as company claims. Zoom also said it expects to pass $5 billion in annual revenue this year (company expectation).
2.2 Product Hunt: what launched and what ranked
According to Product Hunt's weekly board for Oct 19 to 25, which was still open on Oct 24, the top office-type launches are the four in the lead table: Beam.win lets you run AI agents on one Mac and use its windows on another, readywhen is an assistant inside WhatsApp, Kai Resolve by Gleap is a support agent that "does the digging", and Wally by Willow is a scribe agent that learns how you write. Quello ("one place to ask across everything you already use", a cross-tool search) ranked 11 with 84 votes.
The page shows two counts per product. We read the second as votes and the first as comments, but the page does not label them. The ordering is not strictly by votes, because rank 9 (90) sits above rank 10 (91). The board is a discovery list. None of these entries shows a maker, price or availability, so nothing here goes beyond "worth a look".
Gleap's own award page confirms Kai Resolve was Launch of the Day on Oct 19, 2026.
For Oct 22, the Product Hunt daily pages returned no entries. We use StartupCorners' digest instead, which is secondary. According to it, the day's launches included Notion Auto (rebuilt agents that run for hours "at significantly lower cost"), UseIt Studio (a design agent that outputs editable project files), TwinMind for Desktop (meeting transcription and dictation on Mac and Windows) and Tend (an assistant for entrepreneurs). The digest gives no prices, makers or cost figures, so Notion Auto's saving is an unverified claim.
2.3 Search and creative: thin in the window, priced mostly before it
Search. Yelp's Fall 2026 release, announced Oct 20, includes more than 30 features. Yelp Assistant can now book tables at tens of thousands more restaurants on iOS, which with existing partners passes 500,000 restaurants. Yelp also said about 10% of Request a Quote submissions came through Yelp Assistant in Q2 2026. A "Check for Me" phone agent is "rolling out in the coming months" in the US, so it is not shipping yet. All of this is from a PR-wire commentary piece, not Yelp's release. Kiwi.com's AI Search is described by Kiwi's own magazine as live-fare, natural-language flight search, also reachable through Claude and ChatGPT connectors. It is "still in beta", with UK/EU availability confirmed in an October 2026 check. The launch date and price of the search itself are not stated.
Creative. The material has no primary-source creative launch dated Oct 17 to 23. Framefields ("open-source code-first video") reached rank 10 on the weekly board with 91 votes. OpenArt's page dated Oct 20 tags Seedance 2.5 and GPT Image 2.5 as "NEW" without dates or prices. The usable prices predate the window. Creative Claw, a ChatGPT app quoted on Sep 27, charges $10 per 1,000 credits, so $10 ÷ 1,000 = $0.01 per credit. A 5-second clip costs 40 credits ($0.40) on MiniMax H3 Max Fast and 267 credits ($2.67) on Seedance 2.5 at 720p. That is 267 ÷ 40 ≈ 6.7× more, or $2.67 ÷ 5 s ≈ $0.53 per second. Seedance 2.5 climbs from 119 credits at 480p to 657 at 1080p, about 5.5×.
For AI-search visibility on Shopify, the Vosaire blog (which sells one such app and says so) checked listing prices on Oct 22: Entitled $4.99 a month, ShopRank from $9.99, Kedra $10, $25 and $79, Yumnaz $49 and $149.
The window's only dated, first-party office launch is Zoom's, and most of it is still a roadmap. Everything else is a ranked listing with no stated price, and search and creative prices in the material predate Oct 17.
3. New products: coding and developer agents
Chapter 2 covered office, search and creative launches; in coding, the window (Oct 17–23, 2026) brought fewer new agents than shared tooling, agent infrastructure and repricing. Earlier items are context.
3.1 This week at a glance
| Item | Key figure | Date | Status | Source |
|---|---|---|---|---|
| Adyen Agent Marketplace (open-source CLI) | 5 resource types, 8 target tools | Oct 22 | Released, window | Adyen |
| Zoom AI Services, MCP, Platform Studio | 15 new services | Oct 20 | APIs "now available", window | Zoom |
| Daytona agent sandboxes | 60 ms sandbox setup | Oct 20 | Shipping, window | Fortune; Daytona |
| Claude Code 2.1.302 | Forked subagents default | Oct 16 (Oct 17 per one source) | Shipping, context | Shattered; Developers Digest |
| Cursor "X Subscription" | Plus $8 to Ultra $200 | Oct 12 | In effect, context | Fello AI |
| OpenAI model cut-off for Cursor | Nov 12 | Reported by Oct 21 | Upcoming, reported | Shattered, citing CNBC |
| VS Code 1.137 / 1.142 | Dev Container agents / canvases | Sep 9 / Oct 14 | Context | Shattered; Tech Insider |
| Product Hunt coding-agent tools | Rank 6, 8, 16 (weekly board) | Week of Oct 19 | Discovery only | Product Hunt |
3.2 Window launches: shared setup, headless platforms, agent sandboxes
Adyen Agent Marketplace was released on Oct 22, 2026. Despite the name it is a command-line sync tool, not a store. Adyen calls it "an open source project built by Adyen engineers" under Apache 2.0 that "isn't an Adyen product or service". It handles five resource types (skills, commands, subagents, hooks, MCP servers) and compiles one definition for eight tools: Claude Code, Codex, Cursor, Devin CLI, Factory Droid, Gemini CLI, Opencode and Windsurf. It pins versions, verifies digests and scans for hidden text, but Adyen warns the checks "don't prove a resource is safe to run" because hooks and MCP servers can execute scripts. It is free; the only usage figure is internal (over 1,000 Adyen developers use AI tools daily).
Zoom announced at Zoomtopia on Oct 20 15 new Zoom AI Services, including Voice Agent, Interpreter and Decisions APIs, and says they are "now available for developers". Its MCP integrations already covered Claude, OpenAI Codex and Meta Muse; Microsoft 365 Copilot is added, and Gemini is "coming soon" without a date (announced). No pricing is given.
Daytona announced on Oct 20 a platform of agent sandboxes covering Linux, Windows, macOS and GPU environments (funding: Chapter 6). Its CEO says a sandbox is configured in 60 ms. Daytona says Writer provisioned 300,000 sandboxes in one month after reaching 20,000 a day on its own platform. Both are company claims; no pricing is given.
Product Hunt is discovery only. According to its weekly board for the week of Oct 19, walkai-talkai ranked 6th, buddylist 8th and Lancetta 16th; StartupCorners lists BAND for Developers, Agent Pets and ilha among Oct 22 launches. No prices or maker pages are given, so we recommend none.
3.3 Terminal agents and IDEs: updates just before the window
Shattered reports four tools shipping in the Oct 12–16 stretch: OpenCode 2.0 (Oct 12, background service with an API and SDK), Codex CLI 0.163.0 (Oct 15, steering an agent mid-response), Claude Code 2.1.302 (Oct 16) and VS Code 1.141 with an Agents window (day not stated). Tech Insider dates 2.1.302 to Oct 17 via Releasebot, and says it brought the claude-plugins-official repository, a Mods system and cloud sessions for eligible Pro and Max users. We follow Oct 16: Shattered and the Agent SDK changelog (TypeScript 0.3.302, which tracks CLI 2.1.302) agree. Anthropic's release note is not in the material, so the plugin and Mods claims stay "according to Tech Insider".
The blueprint's VS Code 1.137 shipped Sep 9, not in the window, and added agents in local Dev Containers, shared sessions, automations and voice mode (per Shattered; release notes not in the material). VS Code 1.142 followed on Oct 14 with preview canvases (Shattered credits canvases to 1.141; both are aggregators, so the version is unresolved), chat that moves across Git worktrees and enterprise agent-permission controls. Sep 9 to Oct 14 is 35 days for five releases, consistent with a weekly cadence (our inference); the material has no 1.143.
Adoption, from one aggregator (Labellerr, citing survey and benchmark data):
| Tool | Used in past year | Terminal-Bench 4.0 |
|---|---|---|
| Claude Code | 65.5% | 64.8% (Opus 5.5) |
| GitHub Copilot | 58.7% | no score |
| Google Antigravity | 16% | 87.6%, model only, not comparable |
| OpenCode | 13.6% | no score |
Codex scored 58.2% at about $1.92 a task. Labellerr also says OpenAI reported 5 million weekly active users for Codex in June 2026; treat it as unverified.
3.4 Pricing: Claude leaves Cursor's plans
On Oct 12, 2026 Cursor (Anysphere, acquired by SpaceX on Aug 14 per Fello AI) moved new customers to The X Subscription, which xAI sells with Grok. Fello AI checked prices against vendor pages on Oct 18:
| Plan | Monthly | Annual | Usage vs Plus | $ per usage unit |
|---|---|---|---|---|
| Plus | $8 | $80 | 1x | $8.00 |
| Premium | $30 | $300 | 4x | $7.50 |
| Super | $100 | $1,000 | 15x | $6.67 |
| Ultra | $200 | $2,000 | 50x | $4.00 |
Unit price is monthly price ÷ multiple (30 ÷ 4 = 7.5). Annual billing equals ten months (80 ÷ 96 = 83%). Plus runs Grok models only; on Premium and above, Claude is billed separately from prepaid credit, and Opus 5.5 costs $4 / $20 per million input / output tokens. Shattered's earlier article calls the upper tiers' prices undisclosed; we follow Fello AI, which cites Cursor's documentation. Claude Pro costs $20 ($17 billed annually) and includes Claude Code; Codex is included from the $8 Go plan (daily.dev, checked Oct 23).
OpenAI will stop supplying models to Cursor on Nov 12 (upcoming, reported; Cursor's bundled OpenAI allowance ends, personal API keys keep working). Cursor's CEO says OpenAI models serve about 5% of traffic, a company claim. Shattered also cites a Cursor blog headline of $2 billion recurring revenue, date not stated. Per QCode's Oct 10 calendar it retires six models on Oct 19 (execution unconfirmed), and AgentMaps lists three more on Nov 16.
Claude is no longer part of any Cursor plan, so Cursor's $30 entry to Claude plus tokens costs more than Claude Pro at $20. The window's genuinely new products are plumbing, not new agents.
4. Important updates to existing products
Chapter 3 covered coding agents and developer tools; the established assistants and enterprise agent platforms here changed differently, through privacy controls, an API break and dated launches rather than new flagship models.
| Item | Key figure | Date | Status | Source |
|---|---|---|---|---|
| ChatGPT Locked Chats | PIN or biometrics; 7 plan types | Oct 20 | Rolling out | OpenAI (via TechCrunch) |
| Claude API 2026-10-21 | 3 Messages API breaks; 7 SDKs + CLI | Oct 21 | Released, opt-in | CellCog |
| Gemini launchpads | 1,000-person engineer workforce | Oct 19 | Announced | Accenture |
| Gemini Task (ex-Spark) | Google AI Pro $19.99/mo or Ultra | Date not stated | Shipping | AI Agents Library |
| Copilot Studio Chat agents | 6 capabilities | Oct 26 | Upcoming | Microsoft Message Center |
| watsonx Orchestrate | Live-traffic judge, 3% default sampling | Oct 22 | GA (judge); preview (gateway) | IBM |
| Meta Muse chat sharing | 6,800 exposed links | By Oct 20 | Paused | Dataconomy |
4.1 ChatGPT and Claude: a privacy control and an API break
ChatGPT (OpenAI), window update. OpenAI announced Locked Chats on Tuesday, Oct 20, 2026, and said it was only starting to roll out (rolling out). A locked chat sits behind a four-digit PIN, or Face ID or Touch ID, in a separate section with its own memory, so new details do not feed regular Memory. It is not encryption: OpenAI says locking does not add end-to-end encryption or change retention, and resetting the PIN permanently deletes the locked content. It covers Free, Go, Plus, Pro, Business, Enterprise and Edu, but not ChatGPT for Healthcare, group chats or meeting-recording chats.
Context, before the window: GPT-6 Sol reached paid tiers on Oct 7 and GPT-6 Luna reached Free and Go on Oct 8 (according to Shattered, citing OpenAI). GPT-5.5 left ChatGPT and Codex on Oct 14, with the API unaffected (OpenAI's Models page, via QCode). A batch of 17 OpenAI API deprecation rows was due on Oct 23; QCode last checked it on Oct 10, so we call it scheduled, not confirmed.
Claude (Anthropic), window update. According to CellCog, reading Anthropic's release notes, Claude API version 2026-10-21 shipped on Oct 21 with new major versions of seven SDKs and the ant CLI that send it by default. Nothing breaks until a team opts in (released, opt-in). On the new version, any request with temperature, top_p or top_k returns a 400 error, even at the default value, and document blocks in assistant turns are rejected. Python is 2.0.0 and TypeScript 1.0.0; they send the key as Authorization: Bearer, which can hide it from gateways that read X-Api-Key, and move to Apache 2.0. On Oct 20 Anthropic added user profiles and beta organization webhooks. Claude Code 2.1.302 is in Chapter 3.
Context on the Claude app: Opus 5.5 arrived on Sep 22 at $4 / $20 per million input / output tokens, and Max 5x and Max 20x carry $100 and $200 monthly API credits since Oct 7 (via MemoryPlugin). Shattered mentions an undated 20% Opus 5.5 price cut; we follow the $4 / $20 list price that two sources give.
Neither vendor shipped a new model in the window. OpenAI shipped a privacy control with stated limits, and Anthropic shipped an API version whose cost is migration work that teams can schedule.
4.2 Gemini, Copilot and watsonx: agent platforms for work
Google Gemini. On Oct 19, Accenture Federal Services and Google Public Sector announced "Gemini launchpads", a hub of forward-deployed engineers (FDEs) to build Gemini agents for federal agencies. Accenture's separate commitment is a 1,000-person FDE workforce, and pre-built HR, finance and supply-chain agents are still in development (announced; no price). In the Gemini app, Spark was renamed Task, now behind a Chat/Task toggle. Task needs Google AI Pro or Ultra, age 18+, and Keep Activity and Personal Intelligence turned on, and is unavailable in the EEA, UK, Switzerland and Nigeria. Google has not dated the change (shipping; Help Center as of Oct 19).
Microsoft Copilot. Copilot Studio's six capabilities for Copilot Chat agents (topics, connected agents, evaluations, monitoring, preview with Activity Map, lifecycle management) reach GA on Oct 26, 2026, two days after this report (upcoming). Business Central 29.0, released in October (day not stated), lists 8 Copilot and agent updates: 7 GA and 1 public preview (according to TMR Consulting). Copilot Chat in shared Outlook mailboxes is in development for November on the GCC High roadmap. Partner Center lists usage-based billing as default for new Microsoft 365 Copilot Business licenses from Dec 1, 2026; only the headline is in our material.
IBM watsonx Orchestrate (October update, Oct 22). The live-traffic LLM-as-a-Judge is GA, with six evaluators, a 3% default sampling rate and overrides up to 20%. IBM's cost example: 10,000 runs a day at 3% is 300 sampled runs, and with six evaluators 300 × 6 = 1,800 judge calls; at 20%, 10,000 × 20% × 6 = 12,000. The AI Gateway is in preview for Microsoft Foundry and Google's Gemini Enterprise Agent Platform.
4.3 Meta Muse, Perplexity and the plan-price ladder
Meta Muse (window update). Meta paused Muse chat sharing after Claroc found 6,800 shared links discoverable through passive DNS, because each link put the conversation title in a subdomain. Dataconomy's URL is dated Oct 21, but 9to5Mac reported the suspension on Oct 20, so we date the pause by Oct 20 (in effect; no restart date). Muse launched on Sep 8 in the US and is "free for most everyday tasks"; Layer3Labs lists $20 Power and $100 Maximum tiers. Download data is in Chapter 5.
Perplexity. Our material has no Perplexity update dated Oct 17-23, so we report none. Its published plans are Pro at $20 a month and Max at $200.
PowerfulAI.tools, checking vendor pages on Oct 20, is the one source that prices the rivals side by side; it has no ChatGPT row.
| Plan (monthly) | Price | Source |
|---|---|---|
| Google AI Plus / Pro | $4.99 / $19.99 | PowerfulAI.tools |
| Claude Pro (monthly billing) | $20 | PowerfulAI.tools |
| Microsoft 365 Premium | $19.99 | PowerfulAI.tools |
| Perplexity Pro | $20 | PowerfulAI.tools |
| Meta One Premium | $19.99 | PowerfulAI.tools |
Five vendors cluster at about $20, so the difference is what the plan unlocks; Gemini Task, for example, needs the $19.99 plan.
5. Users and revenue data
Chapter 4 covered what changed inside established assistants; this chapter asks how many people and how much money stand behind them, and who disclosed it.
| Item | Key figure | Date | Status | Source |
|---|---|---|---|---|
| Meta Muse, US downloads | 10M+ in first 40 days (12th app ever) | Oct 20, 2026 | Tracker estimate, reported | Sensor Tower via 9to5Mac |
| Meta Muse, retention | 30% day 7, 23% day 14 (peers 19% / 16%) | Oct 20, 2026 | Tracker estimate, reported | Sensor Tower via 9to5Mac |
| Meta Muse, all downloads | 10.9M in six weeks, 10.2M of them US | Oct 20, 2026 | Reported (promotional commentary) | Sensor Tower via Yahoo Finance via PR Newswire |
| Podium, AI ARR | Over $150M in 2.5 years | Oct 22, 2026 | Company statement | Podium (PR Newswire) |
| KnowBe4, AI-related ARR | Up 35%+ in 2026 so far; base not disclosed | Oct 21, 2026 | Company statement | KnowBe4 (Business Wire) |
| Daytona, revenue run rate | About 25x Jan to Sep; base not disclosed | Oct 20, 2026 | CEO claim, reported | Fortune |
| ChatGPT apps directory | 4,200+ apps; no usage data published | Query run Oct 20, 2026 | Third-party census | BigIdeasDB |
| Tolan, users and ARR | 250,000+ monthly users; ARR more than doubled | Oct 19, 2026 | Company figures, reported | TechCrunch via Global Dating Insights |
| Character.AI, users and revenue | About 12M monthly users; about $5.3M revenue in Aug 2026 | Filed Oct 13, 2026 (context) | Sworn declaration, secondary | Afterlife.ai |
| AI vs non-AI apps, annual retention | 21.1% vs 30.7% | Date not stated | Industry benchmark, secondary | RevenueCat via Creem |
5.1 Meta Muse: the window's best-measured product
Meta's Muse launched on Sep 8, 2026, so it is a context product; the window item is new data about it. Muse is a personal agent that runs errands such as travel, forms and shopping from a chat thread, using its own cloud computer and browser. It is free for most uses with subscription plans, and rollout is US-only on iOS, Android, muse.ai and WhatsApp. MediaPost reported business-specific features at up to $100 a month on Oct 1. Meta's own user count is not in the material, so every figure below is Sensor Tower's estimate relayed by media.
| App | US downloads in first 40 days |
|---|---|
| Meta Muse | 10M+ |
| ChatGPT | 3.7M |
| Grok | 0.5M |
| Claude | 0.2M |
Sensor Tower via 9to5Mac, Oct 20, 2026.
On a US basis, Muse is 10 ÷ 3.7 ≈ 2.7x ChatGPT's launch pace. On a global basis it is not: Sensor Tower puts ChatGPT at about 15M global downloads in its first 40 days, against Muse's 10.9M after six weeks, 10.2 ÷ 10.9 ≈ 94% of them in the US. Retention is the better signal. US Muse users returned at 30% on day 7 and 23% on day 14, while ChatGPT, Claude and Grok averaged 19% and 16%. Roughly three in four Muse users (100 − 23 = 77%) were still gone by day 14. Sensor Tower also estimates about 2.5M daily users, first reached on Oct 16, up 15% week on week; the material does not say whether that is global or US.
The growth was bought. Muse took 39% of Meta's US advertising spend in October so far, up from 23% in September. Gmelius, citing Sensor Tower, says downloads rose 73% in one day after the ad push began on Sep 16. Apptopia data, as relayed by Gmelius, show more than 95% of Muse users also use Facebook.
Muse is the 12th app ever to reach 10M US downloads in 40 days, and Meta is paying heavily to drive it: its proven edge is retention above peers, not reach beyond one country.
Three conflicts. Garbage Day says Muse has "nearly 10 million" downloads across both stores; we follow Sensor Tower's 10.9M. Garbage Day also reads day-14 retention as "around 20%"; we follow 9to5Mac's 23%. Chart rank is thin evidence: Appfigures data show a restaurant app beat Muse by about 7,000 downloads on Oct 13. A Deutsche Bank analyst estimate puts Muse at up to 8% of Meta revenue by 2030, about $36B, with a $2.4B low case. That implies Meta revenue of 36 ÷ 0.08 = $450B. Separately, researchers found shared Muse conversations could expose sensitive data, and Meta suspended sharing (by Oct 20, as in Chapter 4).
5.2 ARR claims: who disclosed what
An ARR claim is only comparable when it names its base. In the window, two companies gave statements and one gave a growth multiple.
- Podium (local-business customer agents; calls, leads and bookings) said on Oct 22 that it passed $150M in AI-specific ARR in two and a half years, versus eight years for its original software, and that 20,000+ agents are deployed. The arithmetic: 8 ÷ 2.5 = 3.2x faster, an average of $60M of ARR a year. Podium also says its customers generated $13B in revenue with its agents, a company claim without a stated method. Price: not stated.
- KnowBe4 (security training and AI-agent security) said on Oct 21 that AI-related ARR grew more than 35% in 2026 so far, on 70,000+ customers. No dollar figure is given.
- Daytona (sandbox computers for agents) says its run rate grew 25x from January to September, from an undisclosed base. The CEO's 50.9% average monthly growth over eight months gives 1.509^8 ≈ 27x, consistent with that. Its funding is covered in Chapter 6.
The large-lab numbers are weaker. MarTech (Oct 15, before the window) flagged OpenAI ARR claims of $50B versus $70B, unverified. Exploding Topics repeats $50B as "October 2026" without a source. WSJ's reading of a draft Anthropic prospectus, relayed by Newsquawk (date not stated), gives H1 2026 revenue of $16.3B. With Q2 at $11.6B, Q1 is 16.3 − 11.6 = $4.7B. A straight-line Q2 run rate is 11.6 × 4 = $46.4B, so the "$110B by year-end" investor expectation needs about 2.4x more. According to shattered.io, Cursor's blog claims $2B recurring revenue, doubled in three months (date not stated).
5.3 Directories, companions and retention
ChatGPT apps directory. BigIdeasDB counted 4,200+ apps on Oct 20, sorted into 24 clusters. The directory publishes no installs, ratings or rankings, and only 12.7% of apps declare tools. Three clusters (AI models and agents, web search, travel) hold 27.0% of apps, while HR, support and CRM hold 11.0%. Jotform says the directory has "over 60" workplace integrations, depending on region and plan. These probably count different things, and we follow 4,200+ for total size, as a third-party census. OpenAI's weekly users are not disclosed in the window: Exploding Topics lists 810M in Nov 2025, and Jotform says "around 900 million", undated.
Companions. Tolan, a voice-first companion, reported 250,000+ monthly users, 70M+ conversations in a year and ARR more than doubled to the double-digit millions, as relayed from TechCrunch on Oct 19. Users have paid $10 to $200 a month, and a first free tier is planned (round in Chapter 6). CellCog calls the 250,000 "active users"; we follow the "monthly" wording. Character.AI's Chapter 11 declaration (filed Oct 13, context) gives about 12M monthly users, $2.8M of subscriptions plus $2.5M of ads (= $5.3M) in Aug 2026, and $56.1M of compute in January to August, 45% of operating expenses. At $5.3M ÷ 12M, revenue is about $0.44 per monthly user. Retention. According to Creem, citing RevenueCat's 2026 report, AI apps beat non-AI apps on conversion but lose on retention:
| Metric (median) | AI apps | Non-AI apps |
|---|---|---|
| Trial-to-paid conversion | 8.5% | 5.6% |
| Monthly realized lifetime value | $18.92 | $13.59 |
| Annual-plan retention at 12 months | 21.1% | 30.7% |
| Refund rate | 4.2% | 3.5% |
Check: 21.1 ÷ 30.7 = 0.69, so about 31% fewer annual subscribers remain. Menlo Ventures (via Garbage Day) says about half of US AI users pay, while a Bank of America report says about 3% of households pay. Half of 64% would be 32%, so the two cannot share a base; treat both as estimates.
6. Funding, acquisitions and pricing models
Chapter 5 sized the audiences and revenue claims; this chapter follows the money into the companies behind them and then to the price list customers actually see.
6.1 Deals and pricing moves in the window
| Item | Key figure | Date | Status | Source |
|---|---|---|---|---|
| Daytona Series B | $60M at $1B, led by FirstMark; total $91M | Oct 20 | Announced by company | Daytona |
| Tolan funding | $30M, led by Union Square Ventures; company valuation undisclosed | Oct 19 | Reported | Global Dating Insights, citing TechCrunch |
| Azoma Series A | $11.2M | Oct 21 | Announced | Techstars |
| Kapital to acquire Woz | $30M, cash and stock | Oct 19 | Agreement announced, closing not stated | Kapital via PR Newswire |
| OpenAI acquires Orchestra Bio | Terms undisclosed | Oct 20 | Reported (Axios exclusive, shared by OpenAI) | Axios |
| Chapter acquires Ridgeway | Terms undisclosed | Oct 22 | Reported | citybiz |
| Oracle AI Agent Studio for Industry Applications | Metered in "AI units", no rates published | Oct 21 | Announced, availability planned | Oracle |
| Per-resolution list prices | $0.50 to $2.00 | Checked Oct 18 | In effect | QueryNow |
Money went to infrastructure and small specialist teams, and most acquisitions disclosed no price. Pricing is settling on seat plus metered credits, with true outcome billing limited to customer support.
6.2 Funding and acquisitions: specialist teams, thin disclosure
Funding, briefly (the funding issue covers the market). Daytona sells cloud "sandbox" computers where agents run code, for AI labs and agent builders. Its $60M Series B ($1B) took total funding to $91M (2 + 5 + 24 + 60). CEO Ivan Burazin says run-rate revenue grew about 25× from January to September, averaging 50.9% a month (company claim; absolute revenue undisclosed). Compounding checks out: 1.509^8 ≈ 26.9. Daytona's other claim does not: about 135% monthly net revenue retention would give 1.35^6 ≈ 6×, not the "nearly three times" it also states. Treat both as unaudited. Pricing and availability terms are not in the material. Azoma, which helps consumer brands appear in AI shopping agents, says revenue grew almost tenfold in a year. Tolan's valuation is a PitchBook estimate of more than $268M, not a company figure. Reduck's $3M pre-seed was reported Oct 22 via shattered.io citing Tech.eu. Agentjesse.ai lists Legora's $550M Series E in "October" with no day, so it cannot be placed in the window. Modal's $750M raise on Oct 14 predates it.
Acquisitions. Kapital, a Mexico City financial group, agreed on Oct 19 to buy Woz, maker of WOZCODE, a token-reduction tool for coding models, for $30M. Woz says it has saved customers over $2.5M, a company claim. Whether WOZCODE continues as a product is not stated. On Oct 20 OpenAI said it had acquired Orchestra Bio, which builds drug-development agents, and will integrate it into its Rosalind life-sciences suite. Chapter bought Ridgeway, which had raised $2.8M, on Oct 22 and did not say whether Ridgeway's products survive. Automation Anywhere (Boost.ai), HCLSoftware (Robotiq.ai) and Barndoor (Diaphora) appear on an Oct 23 news page with no deal dates.
Acquirers are buying engineers, not products. For founders, an acqui-hire is the realistic exit for a small agent startup. For users of an acquired product, continuity is unconfirmed in all three cases.
6.3 How agents are priced: the meter decides the bill
Oracle announced Oracle AI Agent Studio for Industry Applications on Oct 21: a no-code and low-code tool for Oracle's finance, retail, construction and government customers to build and run agents, with MCP support. It lets users monitor "AI-unit consumption", but no rate is published. FF News calls it launched. Oracle says availability is "planned within the next 12 months", US first, and we follow Oracle. Sia AgentLaunch is not in the material.
QueryNow read these vendor pages on Oct 18:
| Product | Billed unit | List price |
|---|---|---|
| HubSpot Customer Agent | Resolved conversation | $0.50 |
| Fin (Intercom) | Outcome, max one per conversation | $0.99 |
| Gorgias AI Agent | Resolved interaction | $0.90 annual, $1.00 monthly |
| Zendesk AI agents | Verified resolution | $1.50 committed, $2.00 pay-as-you-go |
| Salesforce Agentforce Help Agent | Resolution, packs of 1,000+ | $2 |
| Agentforce Flex Credits | Action (20 credits) | $500 per 100,000 credits |
| Copilot Studio | Credit (answer 2, action 5) | $0.01 pay-as-you-go |
Flex Credits work out at 500 ÷ 100,000 × 20 = $0.10 per action. A Copilot Studio generative answer costs 2 × $0.01 = $0.02 pay-as-you-go. QueryNow calls Fin "now part of Salesforce"; other sources do not, so we omit it. Sierra, Decagon, Ada and Crescendo price per outcome or conversation but publish no rate.
The unit matters more than the rate. Definitions differ: Gorgias bills a normal ticket if a person replies within 72 hours, and Fin deducts an outcome if the customer returns. Credits do not roll over at Salesforce. HubSpot halved its Customer Agent price from $1.00 on Apr 14 (context).
Seats have not gone away. Helply, a vendor that ranks itself first, priced five agents, 1,000 conversations and 500 AI resolutions at Oct 20 annual list prices:
| Tool | Monthly total | Seat share |
|---|---|---|
| Helply | $339 | Unlimited users |
| Help Scout Plus | $600 | 38% |
| HubSpot Service Hub | $670 plus $1,500 once | 67% |
| Intercom Essential | $785 | 18% |
| Zendesk Suite Professional | $1,500 | 38% |
Zendesk costs 1,500 ÷ 339 ≈ 4.4 times the cheapest option, and seats are 575 ÷ 1,500 ≈ 38% of it. In a survey cited by one blog, seat-based pricing fell from 21% to 15% of companies in a year while hybrid rose from 27% to 41%. Per HubSpot partners, EMEA new customers moved on Oct 1 to credit pools, with 50 credits per resolved conversation. On coding tools, GitHub Copilot moved to usage-based AI Credits on June 1, according to amux. Cursor's bundled OpenAI allowance ends on Nov 12 and OpenAI models then bill at API list price, reported by shattered.io (see Chapter 3); Claude is already billed outside Cursor's plans since Oct 12. According to CellCog, Tolan users have paid $10 to $200 a month, and Tolan plans its first free tier.
The judgement: compare vendors on cost per finished task, not sticker price. Gartner estimates generative AI cost per customer-service resolution will pass $3 by 2030 (via QueryNow), so even a low per-result price can rise, and contracts should say when and how.
7. Try-this list
Chapters 2 to 6 covered launches, updates, user and revenue figures, and money. This chapter turns them into a reader-by-reader table, a short try-list and a dated watch-list. It uses only facts from those chapters.
7.1 What changed, and what it means by reader
| Reader | What changed | Implication | Evidence |
|---|---|---|---|
| Users | ChatGPT Locked Chats (PIN or biometrics) rolling out from Oct 20, with no added end-to-end encryption; Meta paused Muse sharing after 6,800 exposed links; Gemini Task is unavailable in the EEA, UK, Switzerland and Nigeria | Use locking as a view control, not as encryption; do not share Muse conversations until sharing returns; check your region before paying | — |
| Builders | Claude API 2026-10-21 returns a 400 error for sampling parameters on the new version; Adyen's marketplace compiles one definition for 8 tools; Zoom opened 15 AI Services on Oct 20 | Schedule the API migration on your own date; share setups across tools but review every hook and MCP server; the new Zoom services are APIs with no stated pricing | — |
| Buyers | Claude sits outside Cursor's plans; per-resolution list prices run from $0.50 to $2.00; Oracle meters in "AI units" with no published rate; usage-based billing becomes the default for new Microsoft 365 Copilot Business licenses from Dec 1 | Compare cost per finished task, and ask for the meter definition in the contract | — |
Most of the week's change sits in controls, migrations and meters, not in new flagship agents, so the useful move is to read the unit and the date before the headline.
7.2 Try-list by scenario
| Scenario | Try | Price | Caveat | Evidence |
|---|---|---|---|---|
| Meetings and office | ZoomMate Agent Teams | Not stated | Only one piece is live; the other six features are dated to Oct 2026 through Q1 2027 | — |
| Multi-tool coding setups | Adyen Agent Marketplace | Free, Apache 2.0 | Adyen says its checks "don't prove a resource is safe to run" | — |
| Agent infrastructure | Daytona sandboxes | Not stated | The 60 ms setup time and the Writer figure are company claims | — |
| Personal agent | Meta Muse | Free for most uses; US only | Sharing is paused; the retention lead is a Sensor Tower estimate | — |
| Private chat | ChatGPT Locked Chats | Included in Free to Edu plans | Not encryption; resetting the PIN deletes the locked content | — |
| Support automation | Kai Resolve by Gleap | Not stated | Discovery listing; Product Hunt entries show no price | — |
7.3 Key dates
Dates already past on Oct 24, 2026 are marked "In effect". Items given only by month are listed after the dated items of that month.
| Date | Event | State on Oct 24, 2026 | Source |
|---|---|---|---|
| Oct 12, 2026 | Cursor moves new customers to The X Subscription | In effect | Fello AI |
| Oct 16, 2026 | Claude Code 2.1.302 ships | In effect | Shattered |
| Oct 20, 2026 | ZoomMate Agent Teams available; Locked Chats begin rolling out | In effect | Zoom; OpenAI via TechCrunch |
| Oct 21, 2026 | Claude API version 2026-10-21 released, opt-in | In effect | CellCog |
| Oct 22, 2026 | Adyen Agent Marketplace released | In effect | Adyen |
| Oct 26, 2026 | Copilot Studio Chat agents capabilities reach GA | Upcoming | Microsoft Message Center |
| Oct 2026 (day not stated) | Zoom My Notes Workflows | Upcoming | Zoom |
| Nov 12, 2026 | OpenAI stops supplying models to Cursor (reported) | Upcoming | Shattered |
| Nov 16, 2026 | Three more Copilot model retirements listed | Upcoming | AgentMaps |
| Nov 2026 (day not stated) | Zoom enriched notes and DTEN My Notes device | Upcoming | Zoom |
| Dec 1, 2026 | Usage-based billing default for new Microsoft 365 Copilot Business licenses | Upcoming | Partner Center |
| Dec 2026 (day not stated) | Zoom Follow the meeting and new Zoom Chat | Upcoming | Zoom |
| Q1 2027 | Zoom Bota My Notes device | Upcoming | Zoom |
8. References
This document draws on the following 76 sources, listed in the order they are first cited.
- Zoom advances its AI strategy with a new Unified Context Layer that helps turn meeting conversations into completed work | Zoom — news.zoom.com
- Best of Product Hunt: Week of October 19, 2026 | Product Hunt — producthunt.com
- Gleap Awards (2026) | Product Hunt — producthunt.com
- One setup for every AI tool: Agent Marketplace - Adyen — adyen.com
- Daytona Raises $60M Series B to Build the Agent Cloud — daytona.io
- Cursor vs Claude Code: Which to Pay For in 2026 — felloai.com
- OpenAI Cuts Cursor Off Nov. 12 Over SpaceX Deal 2026 — shattered.io
- ChatGPT now lets you lock your chats — techcrunch.com
- Anthropic API 2026-10-21: Breaking Changes and SDK 2.0 | CellCog — cellcog.ai
- Meta Pauses Muse Sharing After 6,800 Links Exposed - Dataconomy — dataconomy.com
- Meta’s Muse app tops 10 million US downloads in 40 days - 9to5Mac — 9to5mac.com
- Podium Surpasses $150M in AI Agent ARR in 2.5 Years | State News | hjnews.com — hjnews.com
- Podium Surpasses $150M in AI Agent ARR in 2.5 Years — finance.yahoo.com
- AI Apps Convert Great But Churn 30% Faster: The Retention Paradox Every Builder Needs to Solve | Creem — creem.io
- Kapital Agrees to Acquire AI Startup Woz for $30M to Enhance Capabilities of AI-Native Banking Platform — prnewswire.com
- AI Agent Pricing 2026: Per Seat, Per Action, Per Result | QueryNow® — querynow.com
- Oracle AI Agent Studio for Industry Applications Helps Accelerate Business Outcomes — oracle.com
- Meta unveils Muse, a personal AI assistant — fox10tv.com
- Best Product Launches: October 22, 2026 — startupcorners.com
- Ten Million Downloads in Six Weeks: The Fight for the AI Agent in Your Pocket Has Started — TradingView News — tradingview.com
- Best AI for Flights in 2026: 9 Tools That Search (and Book) Flights | Kiwi.com Magazine — kiwi.com
- Cheapest AI Video Generator in ChatGPT (2026 Cost per Clip) — creativeclaw.co
- Best of Product Hunt: October 22, 2026 | Product Hunt — producthunt.com
- Best of Product Hunt: October 22, 2026 | Product Hunt — producthunt.com
- Best AI Video Generators in 2026 — Tested & Compared — openart.ai
- Best AI Visibility Apps for Shopify (2026) | Vosaire — vosaire.com
- Zoom broadens its developer platform with new AI Services, MCP expansion, and a unified platform studio | Zoom — news.zoom.com
- Exclusive: Daytona, a startup building the ‘agentic cloud,’ valued at $1 billion in funding round | Fortune — fortune.com
- Claude Code, Codex CLI, OpenCode Update in 4 Days — shattered.io
- Claude Agent SDK in Late 2026: What Is New and How to Use It — developersdigest.tech
- VS Code 1.137 AI Agents in Dev Containers 2026 — shattered.io
- VS Code 1.142 Update: Agent Canvases, Python Fixes 2026 — tech-insider.org
- Claude Code Plugins Go Official at v2.1.302 2026 — tech-insider.org
- 10 Best AI Coding Agents in 2026 — labellerr.com
- Claude Code vs Codex (2026): Which One to Pick | daily.dev — daily.dev
- Cursor vs Claude Code: $8 vs $20 Entry Plans 2026 — shattered.io
- AI Model Releases October 2026: GPT-5.5 Leaves ChatGPT — qcode.cc
- Timeline — AI agent launches, funding & acquisitions — AgentMaps — agentmaps.org
- Accenture Federal Services and Google Public Sector Establish Gemini Launchpads to Power Purpose-Built Agents Across Federal Missions — newsroom.accenture.com
- Gemini Chat Vs Task (Formerly Spark): What's the Difference? — aiagentslibrary.com
- MC1518233: Copilot Studio - Enhanced support for Copilot Chat agents — msmessagecenter.com
- What's New in watsonx Orchestrate: October 2026 Update for Builders — community.ibm.com
- GPT-6 Intelligent UI Rollout in ChatGPT Explained (2026) — shattered.io
- Claude New Features Oct 2026: Opus 5.5, Sonnet, Haiku — blog.memoryplugin.com
- Google Cuts Free Gemini to 1 Model After Argon Launch — shattered.io
- How to Choose the Best AI Chatbot in 12 Steps 2026 — shattered.io
- Microsoft Dynamics 365 BC 2026 Wave 2 Copilot Updates: The Definitive Guide - TMR Consulting Private Limited — tmrconsult.com
- GCC High Microsoft 365 Roadmap Updates - October 22, 2026 - Planet Technologies — go-planet.com
- October 2026 announcements - Partner Center announcements — learn.microsoft.com
- Meta Muse Alternatives: Top AI Agents Compared (Tested 2026) — layer3labs.io
- Best ChatGPT Alternatives in 2026 (Free and Paid) - PowerfulAI.tools — powerfulai.tools
- KnowBe4 Delivers Strong 2026 Momentum As It Secures the Entire Workforce, People and AI Agents — finance.yahoo.com
- KnowBe4 delivers strong 2026 momentum as it secures the entire workforce, people and AI agents | ZAWYA — zawya.com
- ChatGPT App Ideas 2026: 17 Gaps in a 4,200+ App Directory — bigideasdb.com
- AI Companion Tolan Raises $30M as Revenue and User Engagement Grow - Global Dating Insights — globaldatinginsights.com
- AI Companion Statistics 2026: Users, Teens, Money | Afterlife AI — afterlife.ai
- Meta Adds Business-Facing Muse AI Features 10/01/2026 — mediapost.com
- Meta Muse and Wonder AI: Everything You Need to Know About Meta’s AI Agents | AI Agents | Gmelius — gmelius.com
- Is Meta's Muse assistant actually popular? — garbageday.email
- The latest AI-powered martech news and releases — martech.org
- Number of ChatGPT Users (2026) — explodingtopics.com
- Oracle (ORCL) launches an AI agent studio for industry applications | Newsquawk — newsquawk.com
- I tested 20+ of the Best ChatGPT productivity apps — these 10 made the cut | Jotform Blog — jotform.com
- Tolan Raises $30M for Its AI Companion: The Numbers | CellCog — cellcog.ai
- Azoma Raises $11.2M Series A to Help Brands Sell Through AI Shopping Agents | Techstars — techstars.com
- Exclusive: OpenAI acquires biotech startup as AI race moves into drug development — axios.com
- Chapter Acquires AI Startup Ridgeway to Expand Technology for Medicare and Retirement Services | citybiz — citybiz.co
- Daytona raises $60M at $1B valuation to build a cloud for AI agents — TFN — techfundingnews.com
- Reduck Secures $3M for AI Agent Web Automation — shattered.io
- Top AI Agent Startups in 2026: Companies & Funding — agentjesse.ai
- Automation Today: Publication & Newsletter: American Express Introduces Framework to Verify Customer Intent in Agentic Commerce — automationtoday.net
- Oracle AI Agent Studio: Revolutionizing Financial Services with No-Code AI Workflows — ffnews.com
- AI Chatbot Pricing 2026: Billing Models and Real Costs — enjo.ai
- Best Help Desk Software for SaaS in 2026: 9 Tools Compared — helply.com
- How to Buy Software When AI Breaks Pricing (2026), guptadeepak.com — guptadeepak.com
- AI Coding Tools Pricing Compared (2026): Claude Code, Cursor, Copilot, Devin & More — amux — amux.io