Grok, Instinct, Poke: the 2026 AI assistant wars explained

Every tech company is racing to build your personal AI assistant. The differences between them matter more than the similarities. Here's what each one actually does and what it means for you.

2026 is the year the AI assistant went from a feature inside a chatbot to a standalone product category. Every major tech company and a wave of startups have launched dedicated AI assistants that don't just answer questions but take actions on your behalf: booking, scheduling, emailing, shopping, monitoring, and coordinating. The chatbot era (2023-2025) was about answering. The assistant era (2026+) is about doing.

Here's the landscape, what each contender does, what it gets right, and what the entire category is missing.

Instinct: the surveillance assistant

What it does: A text-based personal assistant with deep access to your digital life: email, messages, screen capture, keyboard input, location. You text it tasks. It handles them. The depth of context makes the assistance feel uncanny: the AI knows your patterns, your preferences, and your history.

What it gets right: The capability is the most impressive in the category. Testers describe it as the closest thing to a human executive assistant, because the depth of context produces responses and actions that feel personally informed rather than generically helpful.

The catch: Perpetual, irrevocable data licence. Screen and keyboard logging. Gmail data retained after access revoked. Email sent without approval. Prompt injection vulnerability. The privacy story overwhelmed the product story within days.

Funding: $250 million at a $2.5 billion valuation. No announced revenue model.

Meta Muse: the platform assistant

What it does: A personal AI agent that handles transactional tasks: booking appointments, buying tickets, filling forms, shopping. Runs in a dedicated secure VM with its own browser. Connects to Google Workspace, Ticketmaster, OpenTable, Stripe. Available through WhatsApp and a dedicated app.

What it gets right: The distribution through WhatsApp (billions of users) gives Muse access to a larger potential audience than any standalone app. The secure VM architecture is a more contained approach than Instinct's whole-device surveillance. The consumer task automation is practical and well-executed.

The catch: Meta's data track record ($17 billion settlement) creates a trust deficit. The $130 billion AI infrastructure investment creates pressure to monetise the data flowing through Muse. The capabilities are oriented around consumer transactions rather than professional knowledge work.

Pricing: Free tier with limits. $20/month and $100/month paid tiers.

Poke: the iMessage assistant

What it does: A personal AI agent that lives inside Apple Messages, the first third-party AI agent approved on the platform. Handles scheduling, reminders, smart home control, health tracking, photo editing, and daily briefings. Messages you proactively rather than waiting to be opened. Acquired by Cognition (makers of Devin) in July 2026.

What it gets right: Zero friction. You don't open an app. You text. The iMessage-native approach means the assistant is wherever you already are. The proactive messaging (the assistant texts you first) is a fundamentally different interaction model: the AI doesn't wait for you to remember to ask. It reminds, suggests, and follows up.

The catch: The Cognition acquisition raises questions about the product's direction (Cognition is focused on AI coding agents, not personal assistants). The pricing is unusual (dynamic pricing based on perceived value, with one user reportedly charged $136,000/month). The reliance on Apple Messages limits the audience to iOS users.

Pricing: Free tier. Pro at $19/month. Ultra at $199/month. Custom enterprise pricing.

Grok: the information assistant

What it does: xAI's AI assistant with real-time access to X (Twitter) data, web search, image and video generation, voice mode, and deep research capabilities. Available as a standalone app and through X.

What it gets right: Real-time information is Grok's genuine edge. Access to live X data means Grok can answer "what are people saying about this right now?" in a way no other assistant can. The reasoning capabilities (Grok 4) are frontier-class. Image and video generation from text prompts is well-integrated.

The catch: Grok is primarily an information assistant rather than an action-taking agent. It answers questions and generates content but doesn't book appointments, manage your calendar, or execute transactions. The X association is polarising. The data practices are governed by xAI's terms, which are evolving.

Pricing: Free with limits. SuperGrok subscription for higher usage.

ChatGPT: the incumbent

What it does: OpenAI's general-purpose AI with memory, browsing, code execution, image generation, and voice. The Study Mode and Atlas agent features have expanded it from a chatbot toward an assistant. Memory builds a persistent profile from your conversations.

What it gets right: The broadest capability set: reasoning, coding, research, creative work, voice, vision. The memory feature produces personalised interactions that improve over time. The largest user base creates network effects (more usage data, faster improvement). ChatGPT Atlas handles some agentic tasks (browsing, booking).

The catch: Memory stores facts, not knowledge. ChatGPT knows your preferences but not your files, your meetings, or your team's decisions. The assistant capabilities are built on top of a chatbot architecture rather than designed from the ground up for action.

Pricing: Free tier. Plus at $20/month. Pro at $200/month.

What the entire category misses

Every assistant above operates on the same model: the AI helps you with tasks in the consumer world (scheduling, booking, shopping, researching). None of them address the knowledge work problem: the 60% of working time spent searching for information, attending unnecessary meetings, and maintaining documentation.

The knowledge work problem requires a different architecture:

Connected tools that integrate with where professional knowledge lives (Slack, GitHub, Google Drive, email, CRM, meetings), not just consumer services (restaurants, ticketing, payments).

Self-writing documentation that generates institutional knowledge from team activity, not just answers from general training data.

Semantic search across your accumulated professional knowledge, not just the public internet.

AI agents that handle professional tasks (competitive monitoring, meeting prep, client follow-ups, documentation maintenance), not just consumer tasks (booking, shopping).

A context warehouse where knowledge accumulates, persists, and compounds over time, making the AI more useful every month.

This is what Fabric builds: the AI workspace for knowledge work, complementary to the consumer assistants that handle personal admin. The assistant wars are about who handles your life admin best. The knowledge work question, which is the larger value creation opportunity, is a different race entirely.

Frequently asked questions

Which AI assistant should I use? It depends on what you need. For personal life admin: Poke (if you're on iOS) or Muse (if you use WhatsApp). For general intelligence and research: ChatGPT or Grok. For knowledge work (finding, organising, maintaining, and acting on professional knowledge): Fabric. Most people will use a consumer assistant alongside a knowledge work tool.

Will one assistant win the whole market? Unlikely. The market is segmenting by use case: consumer admin (Muse, Poke, Instinct), general intelligence (ChatGPT, Claude, Grok), and knowledge work (Fabric, Glean). Each segment has different architectural requirements.

Are these assistants safe to use? The safety varies dramatically. Check the data licence (perpetual? irrevocable?), the retention policy (what happens when you leave?), the business model (is your data the product?), and the track record (how has the company handled data before?). Fabric's bring-your-own-storage is the most architecturally private option for professional data.

What about Apple Intelligence and Google Gemini? Both are adding assistant capabilities to their platforms. Apple's on-device approach is privacy-forward but capability-limited. Google's Gemini Agent is strong within the Google ecosystem. Both are platform plays that work best for users already committed to their ecosystem.

Is the text-message interface the future? For consumer assistants, possibly. Poke's iMessage integration and Muse's WhatsApp integration demonstrate that messaging is a natural interface for simple tasks. For knowledge work (complex queries, document management, team collaboration), a dedicated workspace interface is more appropriate.

How much will I spend on AI assistants in total? A reasonable 2026 stack: one consumer assistant (free-$20/month), one general AI (free-$20/month), one knowledge work tool ($10-20/month). Total: $10-60/month. The combined value (hours saved on admin, meetings, research, documentation) typically exceeds the cost within the first week.

Will these assistants eventually merge into one product? Possibly. The consumer assistants (Muse, Poke) could add knowledge features. The knowledge workspaces (Fabric) could add personal admin features. ChatGPT is already trying to be everything. The most likely outcome is continued specialisation: consumer assistants handle life admin, knowledge workspaces handle professional work, and general AI handles everything in between. Specialised tools connected through MCP beat monolithic tools that do everything adequately and nothing exceptionally.

What about enterprise? Do companies adopt these consumer assistants? Consumer assistants (Instinct, Muse, Poke) are designed for individuals. Enterprise adoption requires security reviews, compliance approvals, and data governance that consumer products rarely satisfy. Knowledge work tools like Fabric with bring-your-own-storage and access controls are designed for enterprise requirements. The consumer assistant war and the enterprise knowledge market are parallel races with different winners.

Related reading: The privacy problem with AI assistants, The shift from chatbots to workspaces, Instinct and the case for a private context layer, What Instinct gets right and misses. Related pages: AI assistant, Private and secure, Agents.