AI Assistant Competitive Landscape | AI Wars of 2026

AI assistants are testing new routes to delegated work. Source-backed analysis from AI Wars of 2026.

<!-- Generated from research/AI-ASSISTANT-COMPETITIVE-LANDSCAPE-2026-09-27.html; source-sha256: 420e021d00c018a1f63020fecd8ef3fa1248246b2fa5f41a44af158cbce91ddb. -->

Market map

AI assistants are testing new routes to delegated work.

The question is whether persistent agents can earn permission to act across files, apps, browsers, devices, and transactions. A second question is whether they act reliably enough for people to return.

  • Broad market map
  • Mixed evidence depth
  • Reviewed Oct 3, 2026

Evidence status This snapshot mixes primary product claims, third-party estimates and unverified map entries. Their presence here does not imply comparable evidence or market leadership.

Assistants now reach huge audiences. Trust is still unproven.

Each figure below counts something different. Figures include weekly users across one company’s products, monthly users of one app, a share of business spending, and downloads in one country. The figures cannot be compared with each other, and none of them shows whether people let an assistant do work for them.

1B+

Weekly active users across all OpenAI products, not ChatGPT alone, as OpenAI stated on Sep 8, 2026.

Source: OpenAI

1B+

Monthly active users of the Gemini app, as Google announced on Aug 11, 2026.

Source: Google

40–44%

Anthropic’s share of enterprise AI spending in two analyst estimates. The two estimates use different methods.

Source: Menlo/Ramp synthesis

2.5M+

Estimated US downloads of Meta’s Muse app by Sep 21, from a third-party analytics firm, not from Meta.

Source: Sensor Tower report

Working hypothesis

Some companies already own a place people use every day: a phone, a search engine, a messaging app, or an office suite. Those companies can put an assistant in front of many people quickly. That makes the assistant easy to try. Whether it beats a standalone assistant depends on what happens next: whether people let it use their information, and whether it finishes tasks reliably. It also depends on whether people come back, whether they stay free to choose another assistant, and whether the company earns money from that use. This map does not settle that question.

Three overlapping product positions.

An assistant may be ambient, work-focused and specialist at once. These are lenses for comparison, not mutually exclusive markets.

01 / AMBIENT

Personal agents

Assistants built into consumer surfaces, devices and social products.

  • Examples: ChatGPT, Gemini, Muse
  • Adjacent surfaces: Alexa+, Siri, Grok
  • Question: do reach and context lead to repeat use?

02 / USAGE-BASED

Work agents

Products combining seat access, usage and longer-running tasks.

  • Examples: Claude, Copilot, ChatGPT Work
  • Buyers: IT, operations and business units
  • Question: can governed execution produce measured value?

03 / SPECIALIST

Vertical & local

Focused workflow products and open-source or local-first harnesses.

  • Examples: calendar, browser, coding and GTM tools
  • Counter-position: local-first control
  • Question: does depth persist when suites add similar features?

Profiles with more detail in this research pass.

These profiles are not a ranking or a complete census. “Reported” tags preserve source uncertainty; previews and announced products are not treated as generally available.

MOMENTUM

Meta Muse

META · LAUNCHED SEP 8, 2026

Big ideaPut a persistent personal agent inside the messaging and social apps people already use. Distribution, memory and delegated action are the advantage.

What it does nowMuse works across mobile, web and WhatsApp, then carries tasks into connected services. In early October 2026, Meta added small-business connectors, an enterprise bundle and open-source Muse Gadgets.

Traction and backingSensor Tower reporting puts downloads above 5M, while third-party reporting puts weekly users above 3M. Meta supplies a global consumer distribution base and is extending Muse into business and hardware surfaces.

  • 5M+ downloads reported
  • 3M+ weekly users reported
  • Small Business + Enterprise
  • Open-source Gadgets

Watch-outMac data access triggered a privacy dispute, and retention beyond the launch surge remains unproven.

Muse Gadgets · Small Business · reported traction

SCALE

ChatGPT

OPENAI

Big ideaTurn the default AI destination into a platform that can finish work, not just answer questions. Consumer habit funds a broader agent and developer ecosystem.

What it does nowChatGPT spans chat, voice, research, files, coding and computer use. Dots adds persistent agents with cloud computers and 4,000+ plugin connections; GPT-6.1 Sol was released while Astra was canceled after reported safety failures.

Traction and backingOpenAI says its product family reaches more than 1B weekly active users, not ChatGPT alone. It also reports more than 5M weekly Codex users, giving the company consumer scale and developer gravity.

  • 1B+ product-family WAU
  • 5M+ weekly Codex users
  • Dots: Pro + Business Premium
  • 4,000+ plugin connections

Watch-outAstra’s cancellation shows that faster agent expansion is still constrained by permission and safety failures.

product-family reach · work and Codex · Dots report · Sol report · Astra cancellation

DISTRIBUTION

Gemini

GOOGLE

Big ideaMake AI an ambient layer across Android, Search and Workspace. Native access to Google surfaces and user context is the advantage.

What it does nowGemini combines chat, live voice and camera, Workspace grounding and the persistent Spark agent. Gemini 4 Argon is now the top model, but only vetted cyber defenders and US government users can access it.

Traction and backingGoogle announced more than 1B Gemini app monthly active users in August 2026. Android, Search, Chrome, Workspace and home devices give it the broadest built-in distribution in the field.

  • 1B+ Gemini app MAU
  • Android + Search + Workspace
  • Argon restricted access
  • Free tier narrows Oct 9

Watch-outThe strongest new model is not broadly available, and tighter free access may slow habit-building.

1B MAU report · Argon announcement · restricted access · plan changes

CONSUMER

Microsoft Copilot

MICROSOFT · PERSONAL

Big ideaMake one Copilot app the main entry point to personal and workplace AI. Microsoft expects Windows and Office distribution to turn a scattered assistant into a daily surface.

What it does nowThe free consumer app still handles chat, voice, images and files. A staged redesign adds Home, Code and Autopilot, with personal and work accounts side by side.

Traction and backingNo reliable consumer usage figure appears in the cited sources. The strongest evidence is Microsoft’s built-in reach across Windows, Edge and its Microsoft 365 subscriber base.

  • Free consumer app
  • Unified app preview
  • Home + Code + Autopilot
  • Windows + Edge distribution

Watch-outMost of the redesign is still a staged Frontier preview, so the announced product is ahead of general availability.

Wave 4 summary · unified-app report

ENTERPRISE

Microsoft 365 Copilot

MICROSOFT · WORK

Big ideaOwn enterprise work by grounding agents in Microsoft Graph and governing them inside the tenant. The seat license starts the relationship; metered execution expands revenue.

What it does nowCopilot works across Word, Excel, PowerPoint, Outlook and Teams, with Agent 365 governance. Everyday AI stays in the user subscription while advanced agents and frontier models draw Copilot Credits.

Traction and backingIndustry reporting puts Microsoft 365 Copilot at about 30M paid seats. Its deeper advantage is Microsoft 365 distribution plus tenant identity, data controls and audit infrastructure already accepted by enterprise IT.

  • 30M paid seats reported
  • Microsoft Graph grounding
  • User license + metered usage
  • Agent 365 governance

Watch-outLayering usage charges on top of seat licenses makes budgeting harder, and several agent experiences remain previews.

reported paid seats · new pricing model · Dots + Agent 365

ENTERPRISE

Claude

ANTHROPIC

Big ideaWin high-value work by pairing strong models with a trust and governance story that enterprises can approve. Claude Code is the wedge; connected knowledge work broadens the account.

What it does nowClaude combines chat, connected apps, long-context analysis and Claude Code. Cowork’s capabilities moved into the main chat, while Code Mods now extend tool calls and UI through JavaScript and TypeScript hooks.

Traction and backingTwo enterprise-spend estimates place Anthropic at roughly 40–44%, though their methods differ. Anthropic also committed $100M to train 10,000 Frontier Deployed Engineers by the end of 2027.

  • 40–44% enterprise-spend estimates
  • Claude Code
  • Connected apps + MCP
  • $100M Frontier Academy

Watch-outCode Mods run with full user permissions and no sandbox, so organizations must treat them as trusted code.

enterprise-spend estimates · Code Mods report · Frontier Academy

REAL-TIME

Grok

xAI

Big ideaPair an assistant with X’s live conversation stream, then extend it into autonomous teammates. Real-time data and rapid releases are the differentiators.

What it does nowGrok offers chat, web and X search, media generation and coding tools. Grok Bot adds cloud-computer teammates; a four-tier X + Grok plan up to $100 per month is reportedly under consideration.

Traction and backingX provides a large built-in audience and a proprietary real-time data feed. xAI is now inside SpaceXAI, but neither current product-level usage nor the proposed pricing plan has official confirmation.

  • X full data feed
  • Grok Bot teammates
  • Team Bots beta reported
  • Pricing overhaul reported

Watch-outEnterprise governance is newer than rivals’, and the reported subscription overhaul remains unconfirmed by xAI.

pricing report · Team Bots report

HOME

Alexa+

AMAZON

Big ideaOwn the household’s ambient assistant and connect intent directly to Amazon’s commerce and fulfillment network. The home, not the chat window, is the primary surface.

What it does nowAlexa+ handles conversational smart-home control, household planning and shopping across Echo, Fire TV, mobile and web. A dedicated Fire TV hub begins rolling out in the US in November.

Traction and backingPrime members receive Alexa+ at no added charge, linking the product to Amazon’s roughly 200M reported members. Echo and Fire TV add an installed hardware surface competitors cannot easily copy.

  • Prime included
  • $19.99 standalone
  • Echo + Fire TV
  • Fire TV rollout: Nov 2026

Watch-outAll-cloud voice processing keeps privacy concerns active, and Alexa+ has little reach into office work.

US pricing and rollout · Fire TV rollout report · voice-data caveat

BETA

Siri / Apple Intelligence

APPLE

Big ideaUse on-device processing and Apple’s hardware integration to make personal intelligence private by default. The strategic asset is the assistant built into the device people already carry.

What it does nowApple Intelligence covers writing, summaries, images and notification features across Apple devices. The Siri AI overhaul is rolling out separately as a waitlisted English beta with personal context and wider app actions.

Traction and backingApple has unmatched OS-level distribution across iPhone, iPad, Mac and Watch, but no verified Siri AI usage figure was found. The product is bundled with eligible hardware rather than sold as a separate subscription.

  • OS-level default
  • Waitlisted English beta
  • Device / region eligibility
  • 12GB tier for some features

Watch-outA late beta rollout and new hardware split mean Apple’s distribution advantage still exceeds its delivered agent capability.

rollout status · 12GB split · settlement context

BROWSER

Perplexity + Comet

PERPLEXITY AI

Big ideaOwn research first, then own the browser surface where an agent can compare, decide and act. Citations build trust while Comet moves the company closer to transactions.

What it does nowPerplexity delivers cited answers, deep research and multi-model access; Comet adds an in-browser assistant and background work. The new Decisions API returns probabilities over fixed choices, while Computer now creates interactive charts.

Traction and backingPerplexity reportedly raised about $200M at a roughly $20B valuation in June 2026. Comet is free worldwide, while paid consumer and enterprise tiers fund higher limits and agent workflows.

  • ~$200M raise reported
  • ~$20B valuation reported
  • Decisions API: $0.04/M input
  • Enterprise $40–325

Watch-outComet has faced reported prompt-injection concerns, and Perplexity still lacks a default OS or messaging distribution channel.

funding report · Decisions API + Computer · Amex workflow perks

LEGACY

Pi

INFLECTION AI

Big ideaWin the personal relationship through warmth and emotional intelligence instead of tools or autonomy. Pi treats conversation itself as the product.

What it does nowPi remains a free voice-and-text companion on mobile and web, with planning and reflective conversation. It has no meaningful autonomous-agent or app-integration roadmap.

Traction and backingAppBrain reported more than 1.1M Android downloads and a September 2026 app update. Inflection remains active, but it shifted its company focus and resources toward enterprise models and APIs after the 2024 Microsoft acqui-hire.

  • 1.1M+ Android downloads reported
  • Free
  • Mobile + web
  • No autonomous agents

Watch-outLegacy status makes Pi a respected companion product without a credible path into the current agent race.

company pivot · app status · privacy ranking

AGGREGATOR

Poe

QUORA

Big ideaLet users buy one pool of compute and choose among many labs, models and creator-built bots. Choice and the marketplace matter more than owning a model.

What it does nowPoe combines official models, media generation and user-created bots across web, mobile and desktop. Creators can build prompt, API and interactive-app bots and participate in monetization.

Traction and backingQuora supplies the product and its creator ecosystem, but no reliable source established current 2026 user, revenue or funding figures. Continued multi-platform availability and active creator monetization are the strongest verified signals here.

  • Free daily compute
  • From $4.99 reported
  • Creator monetization
  • Web + mobile + desktop

Watch-outPoe owns neither a foundation model nor an agent runtime, leaving differentiation dependent on aggregation and marketplace liquidity.

profile and reported pricing

SPECIALIST

Lindy

LINDY

Big ideaMake business automation feel like hiring a teammate instead of building a workflow. Natural-language setup and human approvals lower the adoption barrier.

What it does nowLindy handles email, scheduling, calls, meetings and CRM work from Slack, iMessage and a Chrome extension. It connects to 1,000+ tools, supports MCP and requires approval for irreversible actions.

Traction and backingThe strongest evidence in the packet is product breadth: 1,000+ reported connections, 40+ built-in teammate skills and phone agents in 30+ languages. Current user and funding figures were not verified.

  • 1,000+ connections reported
  • 40+ built-in skills
  • 30+ phone languages
  • Plus $29.99/user reported

Watch-outIts workflow depth is valuable, but Microsoft, OpenAI and Google can bundle adjacent automations into products customers already buy.

September product report

VERTICAL

Reclaim + Motion

CALENDAR AGENTS

Big ideaMake the calendar the operating system for personal work. The agent earns trust by protecting focus time and continuously rebuilding the plan as priorities move.

What it does nowReclaim 2.0 adds an AI Assistant, Preview Mode, agentic automations and MCP around Google and Outlook calendars. Motion combines calendar and task constraints into dynamic daily plans that adjust when meetings change.

Traction and backingDropbox acquired Reclaim in 2024, giving it a larger productivity platform behind the focused calendar wedge. No reliable source established current user scale, funding or comparable 2026 traction for Motion.

  • Google + Outlook
  • Dynamic planning
  • Reclaim Preview Mode
  • MCP support

Watch-outDeep scheduling specialization is defensible, but suite-native assistants can absorb the same planning layer.

Reclaim 2.0

VERTICAL

Bardeen

BARDEEN

Big ideaAutomate work directly in the browser when an API is missing or too slow to integrate. The current commercial wedge is go-to-market execution.

What it does nowThe Chrome extension scrapes pages, enriches contacts and qualifies leads for sales teams. Magic Box turns plain-language instructions into automations, while an AI research agent supports prospecting.

Traction and backingCurrent coverage lists a permanent free tier with 100 monthly credits and reported $10 and $50 paid tiers. No reliable user, revenue or funding figures were established, so adoption scale remains unclear.

  • Free: 100 credits
  • $10 / $50 reported
  • Chrome extension
  • No-API automation

Watch-outChrome-only automations can break when sites change, and the sales-automation market is crowded.

profile and reported pricing

INSURGENT

OpenClaw

OPEN SOURCE · MIT

Big ideaKeep the agent, its memory and its tools under the user’s control. Local files and open infrastructure are the counter-position to closed cloud assistants.

What it does nowOpenClaw runs a local-first personal agent across messaging channels, browsers, files and skills. OpenClaw Enterprise adds a free control plane for multi-tenancy, sandboxing, permissions, governance and audit records.

Traction and backingA March 2026 report put the open-source project at roughly 340K GitHub stars. The enterprise effort lists OpenAI, Nvidia, Red Hat and GitHub among sponsors or contributors, while the independent foundation governs the project.

  • Free / MIT
  • ~340K GitHub stars reported
  • Enterprise control plane
  • Local-first

Watch-outTechnical setup and broad local permissions raise the support and security bar for mainstream users.

OpenClaw Enterprise · release report

EU

Mistral Vibe (formerly Le Chat)

MISTRAL AI

Big ideaOffer a European, enterprise-ready alternative that unifies conversation, workplace execution and coding. Regional trust and open tooling are part of the product strategy.

What it does nowVibe is one product with Chat, Work and Code modes across web, mobile, terminals and editors. It connects to Workspace, Outlook, SharePoint, Slack, GitHub and Notion, with plan review before execution.

Traction and backingMistral brings substantial company backing and European enterprise partnerships, but no current Vibe-specific user figure was verified. Its reach is supported by free, Pro, Team and custom enterprise plans.

  • Free / Pro $14.99
  • Team $24.99/user
  • Chat + Work + Code
  • Web + mobile + terminal + editors

Watch-outVibe has a clear regional position, but it lacks the consumer distribution of Android, Windows, WhatsApp or ChatGPT.

Mistral platform FAQ · positioning and pricing

PIVOTED

You.com

ENTERPRISE API INFRASTRUCTURE

Big ideaSupply search, research and data infrastructure to other agents instead of fighting for a consumer assistant habit. The product wants to be a highway, not the destination.

What it does nowYou.com sells Search, Research, Data and Smart APIs plus custom agents. ARI provides deeper research for regulated and enterprise use cases.

Traction and backingThe company raised a $100M Series C led by Cox Enterprises at a reported $1.5B valuation. It reports more than 1B API queries per month, with customers including DuckDuckGo, Windsurf, Harvey and Databricks.

  • $100M Series C
  • $1.5B valuation reported
  • 1B+ API queries/mo reported
  • Enterprise focus

Watch-outThe pivot strengthens infrastructure relevance but removes You.com from the direct consumer-assistant contest.

pivot and funding · infrastructure strategy · round and customers

SHUT DOWN

Phind

ENDED JAN 16, 2026

Big ideaGive developers a search-first answer engine tuned for code and technical documentation. It was a specialist knowledge tool rather than a general agent.

What it does nowNothing: the original Phind.com shut down on January 16, 2026. Pro subscribers received prorated refunds and chat-history export remained available through January 30 before servers were wiped.

Traction and backingNo current traction applies because the product is closed. It remains here only as market history and must not be confused with an unrelated site now reusing the Phind name.

  • Closed Jan 16, 2026
  • Refunds issued
  • Servers wiped Jan 30
  • Not a current competitor

Watch-outThe historical name now creates status confusion; this card tracks the discontinued original product.

closure record

MOMENTUM

Manus

BUTTERFLY EFFECT · 2.0 SEP 28, 2026

Big ideaBuild the best general-purpose execution harness on top of other companies’ foundation models. Manus expects orchestration, tools and the cloud computer to become the durable layer.

What it does nowManus 2.0 rebuilt the agent around Cascade, Cloud Computer, invite-only Cue and Studio. The Singapore-headquartered company now spans research, code, websites, presentations, video, games and event-triggered automations.

Traction and backingButterfly Effect says it passed $100M ARR within eight months, and press reports millions of users. Benchmark led a $75M Series B. Meta’s reported ~$2B acquisition attempt was blocked. Tencent became the largest external shareholder. A reported $500M raise at about $4B remains preliminary talks.

  • $100M+ ARR · vendor-disclosed
  • Millions of users reported
  • $75M Series B · Benchmark
  • 23.2% fewer tokens · Manus says, uncorroborated

Watch-outSalt Labs found a connected-account prompt-injection issue that Manus patched; model dependence and the still-open funding round leave the long-term advantage unclear.

official launch posts · Cascade and 2.0 details · patched security issue · deal and funding talks · traction and backing

The expanded field, by front.

The shared market map covers the categories below. Names marked “map coverage” are included for scope, but their late-2026 status, capabilities and pricing were not independently confirmed in the research packet.

China consumer assistant-app candidates

STARTING LIST · UNEVEN EVIDENCE

  • DeepSeek
  • Alibaba Qwen / Qianwen
  • Moonshot Kimi
  • Baidu Wenxin / ERNIE
  • Tencent Yuanbao
  • ByteDance Doubao

This is a candidate set of user-facing general assistants, not an exhaustive census or ranking. It mixes standalone apps with products linked to wider model and super-app ecosystems; those layers need separate labels.

Market-data limit: QuestMobile's own report dates its category total and app figures to June 2026; earlier news summaries dated the category total to May. The report does not say how users of more than one app are counted, and a second data provider reports different app sizes. Do not convert the figures to shares or a leadership ranking. See the China evidence note below.

Device & ambient layer

MIXED DEPTH

  • Samsung Galaxy AI / Bixby
  • Apple Siri
  • Amazon Alexa+
  • Google Gemini
  • Muse on AI glasses · planned

Default placement may reduce the friction of trying an assistant. This map does not compare reach, repeat use or the value of that distribution.

Samsung is map coverage only. Alexa+, Siri and Gemini are profiled above; Muse glasses support was reported as planned, not released.

Coding-agent harnesses

MIXED DEPTH

  • GitHub Copilot agents
  • OpenAI Codex
  • Claude Code
  • Cursor
  • Windsurf
  • Devin
  • Google Antigravity
  • Replit Agent
  • v0
  • Bolt.new
  • Lovable
  • Aider
  • Cline
  • OpenCode

Several products now offer delegated coding across repositories and development tools. This map does not establish a broader shift or compare task outcomes.

Report-confirmed: Codex, Claude Code, Gemini CLI/Antigravity and Microsoft Code on GitHub Copilot technology. The remaining names are market-map entries with 2026 details open.

Enterprise agent platforms

MAP + REPORT EDGE

  • Salesforce Agentforce
  • ServiceNow Now Assist
  • SAP Joule
  • Workday
  • UiPath
  • Glean
  • Moveworks
  • Kore.ai
  • Amazon Q

This front competes on governed access to enterprise systems of record, workflow execution and auditability — not consumer mindshare.

The report names Salesforce Agentforce, ServiceNow AI and Amazon Q as relevant but outside its consumer scope. All other product specifics remain open.

AI-native browsers

MIXED DEPTH · NO RANKING

  • Perplexity Comet
  • Dia
  • Opera Neon
  • Brave Leo
  • SigmaOS Airis
  • Fellou
  • OpenAI Atlas · ended

The browser is a plausible transaction surface because it can see pages, compare options and act across sites.

Correction: OpenAI discontinued Atlas on Aug 9, 2026; it is not a current browser product. Comet is profiled above. The other entries remain map coverage only.

Workspace-native assistants

MAP COVERAGE

  • Notion AI
  • Slack AI
  • Atlassian Rovo
  • Zoom AI Companion
  • Canva AI
  • Adobe AI

These products defend the application layer by placing assistance directly inside the document, conversation, meeting or creative workflow.

Open questions: current agent autonomy, cross-app reach, pricing, packaging and 2026 launches were not established in the research packet.

Automation & productivity

REPORT COVERAGE

  • Lindy
  • Reclaim
  • Motion
  • Bardeen
  • Zapier Agents
  • n8n
  • Gumloop
  • Granola
  • Shortwave

One question is whether specialist setup and workflow depth can remain distinct as larger suites add related features; this map does not compare those outcomes.

The report profiles Lindy and Bardeen in detail, Reclaim and Motion directionally, and lists the wider field without verified pricing.

Aggregators, companions & infrastructure

REPORT COVERAGE

  • Poe
  • Pi
  • You.com · pivoted
  • Phind · closed
  • Mistral Vibe · formerly Le Chat
  • OpenClaw

Some products aggregate models, some optimize for emotional rapport, and some move toward infrastructure or open-source control.

Phind shut down Jan 16, 2026. You.com is now enterprise API infrastructure rather than a consumer-assistant rival.

Report-backed detail Shared-map breadth; details open

Primary comparison.

Prices are monthly unless stated. “Reported” means the research found only secondary support; “unclear” preserves conflicting sources.

PlayerAutonomy*EcosystemSurfacesPriceMain advantage
MuseHighMeta graph; Plaid, Duffel, Stripe Link; GadgetsiOS, Android, muse.ai, WhatsApp, MacFree; $20/$100 reportedDistribution + persistence
ChatGPTHighWork connectors; Dots agents; GPT-6.1 SolWeb, mobile, desktop, extensionPlan-based + usageScale + complete stack
GeminiHighGoogle data + WorkspaceAndroid, Search, Chrome, Home, XRFree cut to Flash-Lite Oct 9; Pro $19.99Native grounding
CopilotMediumBing, Windows, EdgeWeb, Windows, Edge, iOS, Android, MacFree; M365 Premium $19.99Super-app overhaul
M365 CopilotHigh / previewMicrosoft Graph, M365, Power PlatformOffice, Teams, web, mobileUSL + usage billingTenant context + governance
ClaudeHigh13 connectors, plugins, MCP; Code ModsWeb, mobile, desktop, API, SlackFree; Pro $20; Max from $100Trust + coding
GrokMed–highX data, APIX, standalone app, APIFree + paid; $100 tier reportedReal-time X signal
Alexa+MediumSmart home + Amazon commerceEcho, Fire TV, app, web; Fire TV hub Nov 2026Prime included; $19.99 standaloneHome + fulfillment
SiriLow–medium / betaApple apps + on-device contextiPhone, iPad, Mac, WatchBundledDefault + privacy
PerplexityMediumSources + publishers + model choice; Decisions APIWeb, mobile, Comet browserFree; $5; $20; $200; Enterprise $40 to $325Citations + browser
LindyHigh1,000+ apps + MCPSlack, web, extension, iMessageNot independently verifiedSimple + compliant
OpenClawHighLocal tools + messaging bridges; Enterprise planeMac/server/WSL2 + mobile companionsFree / MITLocal-first control
Mistral Vibe (formerly Le Chat)Med–highWork apps + GitHubWeb/mobile chat; terminal/editors codeFree; Pro ~€14.99 reportedEuropean sovereignty

Comparison scope favors products with more report-backed detail. *Autonomy labels are author impressions without a shared rubric; they are not comparable scores. The broader field remains visible above without invented values.

Questions shaping the race.

These are hypotheses about where competition may develop, not standings or conclusions about who leads.

Personal relationship

The page records product-family WAU, app MAU, and third-party download estimates. Those are different scopes and units; they do not establish which assistant leads in repeat use or user trust.

Enterprise control

Microsoft suite integration, Anthropic spend-share estimates, and OpenAI work products describe different kinds of position. The available evidence does not compare governed task completion, retention, or customer value.

Browser & checkout

Comet and other browser assistants make comparison, form-filling, and transaction handoff worth tracking. Availability does not establish successful or repeated transactions.

Harness layer

Coding agents are a visible test environment for delegated execution. Their availability and coding use do not by themselves establish general work-agent reliability; local-first tools represent a different control model, not a measured counter-winner.

MOMENTUM

Instinct

SPEAR STREET TECHNOLOGY · INVITE-ONLY AUG 2026

Big ideaThe personal assistant should act through the tools you already use. Its position: a text-or-call agent that finishes tasks beats any new app you must learn.

What it does nowNo app, no new interface: you text or call Instinct. It connects to email, calendar, messaging, screen, audio and location, then books travel, pays bills, cancels subscriptions, orders groceries and places calls. It acts with its own phone number and its own computer. On Oct 5, 2026 it added group chats that work even for friends without an Instinct account. Voice calls handle offline bookings, and the Trusted Person Network lets one user's agent coordinate with a friend's agent.

Traction and backingSpear Street Technology has raised about $1.35B. Its Sept 28, 2026 Series C brought $1B at a $10B valuation from Sequoia, Benchmark and Coatue. The Information reports 100K+ invite-only users. Shinn says the platform nears $1B in yearly transaction volume. His growth figures are unverified.

  • $1B Series C · $10B valuation
  • 100K+ users reported
  • ~$1B volume · Shinn says, unverified
  • SMS-only · no app

Watch-outTerms grant a perpetual license to user materials and let Instinct bind users in transactions. Testers reported data retained after disconnecting Google, and one email sent without confirmation. Its Sept 29 "Instinct Selections" recommendations drew public backlash for unsolicited pushes. SMS-only reach leaves it exposed against app-based rivals.

Series C raise · group chat launch · privacy scrutiny · selections backlash

Six signals to watch

  1. Muse retention and paid conversion after launch momentum.
  2. Enterprise acceptance of metered agent compute.
  3. Apple’s path from beta Siri to full modernization in 2027.
  4. OpenAI ads and Sponsored Agents versus answer neutrality.
  5. Which platform captures agentic commerce and checkout.
  6. Safety rules after the first widely visible agent-caused harm.

Sources, limits and corrections.

Verification was rerun Oct 3, 2026. The Oct 3 refresh used news and press sources; single-source items are marked where they appear. First-party pages anchor the highest-confidence claims; index and press sources are retained only where official pages were unavailable.

Corrections that change the map

OpenAI Atlas: officially discontinued Aug 9, 2026; its browser-agent work moved into ChatGPT and Codex. Claude Cowork: Anthropic announced a staged merge with Claude chat on Sep 16, beginning with Pro and Max over the following weeks. Team and Free were to follow, and Enterprise administrators would receive advance notice. Do not describe this as an immediate end for all plans. Primary announcement. Phind: shut down Jan 16. You.com: pivoted to enterprise AI infrastructure. Mistral Le Chat: renamed Vibe. Copilot Pro: replaced by Microsoft 365 Premium for individuals.

Primary-source verified Muse, ChatGPT Work, Atlas, OpenAI and Gemini scale, Microsoft pricing, Claude pricing, Alexa+ pricing.

Third-party estimates Muse downloads and Anthropic enterprise-spend share remain directional.

Not independently verified Exact Grok and Lindy prices; Comet Android launch date.

Values that remain unconfirmed or conflicted

Muse’s $20/$100 tiers are secondary reports. Google Spark availability outside the US remains unclear. Grok and Lindy pricing were not independently verified and are intentionally omitted. Comet’s Android date conflicts across sources. OpenAI’s Sep 8 page states that its products reach more than one billion weekly active users. That is the company’s figure for all its products, not ChatGPT alone, and it counts weekly users rather than monthly ones. Muse downloads and Anthropic enterprise-spend estimates come from third-party measurement and are not comparable to app MAU. The Oct 3 refresh added: the reported X+Grok $100 tier (Bloomberg, unannounced), Anthropic IPO figures (leaked, unverified), and the Alexa+ Fire TV hub date (Amazon announcement).

China assistant-app candidates and agent pathways

Candidate app scope: DeepSeek, Qwen/Qianwen, Kimi, Baidu Wenxin/ERNIE, Tencent Yuanbao, and ByteDance Doubao are starting names for mainland-China-facing consumer assistants. This is not a complete list, a common capability class, or a ranking. Official pages or store listings for each of the six identify the product and its publisher. That establishes identity, not use, reliability or outcomes.

Adjacent agent routes: QuestMobile's July report separates four kinds of agent. The first kind is ecosystem task agents: Qwen task assistant, WeChat XiaoWei, and Alipay Abao. The second is office-delivery agents: Tencent WorkBuddy and Zhipu AutoClaw. The third is device-system agents: Huawei XiaoyiClaw, Xiaomi miClaw, Honor YOYOClaw, and OPPO and vivo equivalents. The fourth is autonomous agents: Youdao Lobster AI and ByteDance TRAE Work. These are QuestMobile's groupings and product descriptions, not verified availability or outcome claims. Keep device, service-ecosystem, desktop, and multi-agent products separate from the standalone-app cluster. One change is confirmed by the company: Tencent's notice says QClaw stops operating on 24 December 2026, with migration to WorkBuddy offered.

Current evidence gap: the book has provisional company records for Baidu, Tencent and ByteDance. A Part II chapter describes six China assistants product by product. Comparable use or outcome evidence is still missing before the cluster can support a book claim. See the China research note for the sources and remaining work.

China usage snapshot: what the figures do and do not show

QuestMobile's half-year report on AI applications (July 14) reports 499 million AI-native-app-category MAU in June 2026, up 85.4% year over year. Category averages were 92.7 uses and 183 minutes per month. It reports 382 million for Doubao, 167 million for Qwen and 130 million for DeepSeek. The report's listed apps together exceed its category total. That is consistent with people using more than one app, but the report does not say how that overlap is counted. IT Home and TechNode summarized the category total as of May; the original dates it to June. A second data provider reports different sizes for the same apps in June and includes products QuestMobile omits. Do not compare the figures or treat them as market shares. Neither frequency nor time spent establishes retention, task success or user value. The report also says Doubao's app added 13.78 million users from May to June; this does not establish the cause or paid conversion.

Keep these names as research candidates, not a verified or exhaustive ranking. iFlytek Spark and Zhipu Qingyan are additional product-screening leads based on reachable official pages. Comparable usage evidence was not found.

Coverage gaps before book incorporation

Search/research: Perplexity appears through Comet, but research assistants such as NotebookLM, Elicit, and professional research tools are not treated as a distinct front. Meetings/communications: Zoom AI Companion appears in the workspace list, but Teams/Meet and specialist meeting/email assistants are not compared. Enterprise and runtime layers: CRM, IT service, ERP, support agents, builders, and model runtimes need clearer separation from end-user assistants.

Cross-cutting control evidence: identity, permission scope, user confirmation, audit, recovery, retention and successful task completion are mentioned unevenly, not assessed on a shared basis. The autonomy labels in the comparison table have no common rubric.

Inclusion rule: distinguish company, model family, app, suite feature, and runtime; set a date, geography, user/task boundary, and evidence-depth label. The saved prior-research inventory is a discovery checklist, not a validated competitor register; its superlatives and unsourced entries are not evidence.

Core source set

A decision-oriented market map, not an investment recommendation. Product availability, pricing and reported scale can change quickly; explicit uncertainty is part of the analysis. Brand marks are shown only to identify the products discussed.

Return to AI Wars of 2026