How to Build an iMessage AI Agent Like SZN

Key Takeaways

• You can build an iMessage AI agent like SZN by combining an iMessage communication layer with an LLM-powered agent backend.
• Start by defining the tasks your agent needs to understand and perform.
  Connect the agent to tools such as web search, calendars, email, or third-party APIs.
• Add memory, authentication, permissions, and safeguards for personalized and reliable interactions.
• Finally, test the agent, deploy it on scalable infrastructure, and continuously monitor its performance.
• Learn how Idea Usher can help businesses develop an iMessage AI agent like SZN with custom AI, integrations, and scalable architecture.

Building an iMessage AI agent like SZN requires a backend architecture that links the messaging layer with an AI engine. A service such as Sendblue or Photon/Spectrum can handle iMessage communication, while a webhook-based backend connects incoming messages to models like Claude or OpenAI, processes the request, and delivers the AI-generated response back to the user.

What makes an agent like SZN interesting is that it goes beyond simply answering messages. It can understand what a user needs, work with external tools, remember context, and help complete tasks without forcing users to switch between multiple apps. In this blog, we’ll discuss how to build an iMessage AI agent like SZN, covering its architecture, key features, integrations, tech stack, development process, challenges, and cost.

Is an iMessage AI Agent a Viable Business in 2026?

Yes, an iMessage AI agent can be a viable business if it runs on an approved channel and completes real tasks rather than simply acting as another chat interface. According to SNS Insider, the AI agents market was valued at USD 7.76 billion in 2025 and is expected to reach USD 316.89 billion by 2035, growing at a 44.9% CAGR. iMessage offers a valuable distribution channel because it is already familiar to iPhone users, although Apple’s platform rules remain a key dependency.

Source: SNS Insider

Growing Demand for Personal AI Agents

Consumer AI use is becoming habitual. Edison Research at SSRS found that 52% of U.S. adults use AI platforms weekly, while Tinuiti found 34% use AI daily and 45% use it more than a year ago. This growing usage creates demand for agents that move beyond answering questions to completing tasks.

Szn illustrates this shift. Built by Nikhil Gupta, a former Siri and Apple Intelligence engineer, it is expected to launch in October as a contact inside Messages, with its own name, email, and phone number. Users can send text, voice notes, photos, or emails, while the agent can browse websites, make purchases, coordinate with vendors, and work with Apple Notes, Find My, and iCloud files.

Demand signals at a glance

• 52% of U.S. adults use AI platforms weekly (Edison Research at SSRS).
• 45% of daily or weekly AI users use AI as a personal assistant (SurveyMonkey, 25,030 U.S. adults).
• 15% of day-to-day work decisions will be made autonomously by agentic AI by 2028, up from 0% in 2024 (Gartner).
• 33% of enterprise software applications will include agentic AI by 2028, up from less than 1% in 2024 (Gartner).

iMessage as a Frictionless AI Interface

The biggest advantage of iMessage is low friction. Tinuiti found that 76% of AI users access AI platforms on their phones. Piper Sandler’s teen survey found 88% of U.S. teens own an iPhone, while Counterpoint reported the average iPhone selling price has surpassed $900. An AI agent inside Messages therefore requires no new app or interface.

Poke demonstrates this model. The Interaction Company of California launched it publicly in March, and on June 4 it became the first AI agent approved for Apple’s Messages for Business platform. It had already relayed about 100 million messages. Apple required live support, clear AI labeling, and compliance with its style guide. Poke also works through SMS, RCS, and Telegram, with WhatsApp available in some markets.

Trust and Platform Access

Trust remains a challenge. Tinuiti found that 48% of people trust AI to recommend products, but only 20% trust it to complete purchases. A verified contact inside Messages can reduce some of that friction. Szn similarly gives its agent a real phone number and email address.

The remaining question is how Szn connects to iMessage. Apple previously blocked Beeper Mini’s unofficial access, while Poke’s approved Messages for Business route is the proven approach described here.

Revenue Opportunities and Market Potential

Several business models are emerging:

Instinct, founded by Noah Shinn and operated by Spear Street Technology, shows the category’s momentum. It is invite-only and works through iMessage, WhatsApp, SMS, and voice, while connecting to email and calendars and handling travel, reservations, and inboxes. Its Trusted Person Network also allows agents to coordinate directly. Its revised terms following criticism over broad data rights highlight why privacy and permissions matter.

What Is SZN? The Ex-Siri Engineer’s iMessage AI Agent Explained

SZN (pronounced “season”) is a personal AI agent that lives inside an iMessage thread and handles tasks such as booking, buying, calling vendors, and following up. Built by former Siri and Apple Intelligence engineer Nikhil Gupta, it gives users a name, email address, and phone number, making it feel like texting a contact rather than opening a chatbot. SZN is currently in free early access, with a wider launch planned for October and a dedicated app to follow.

Built by an Ex-Siri Engineer

SZN was created by Nikhil Gupta through Nomadic Futures, Inc. Bloomberg’s Mark Gurman first reported the project. His background at Apple gives SZN a notable position: an assistant designed to feel native to the iPhone without being built by Apple. Digital assistants are already familiar. 

A Northeastern University and Gallup survey found that 47% of Americans used smartphone assistants such as Siri or Google Voice. However, traditional assistants largely operate within the phone, while SZN focuses on interacting with the outside world through its own phone number and email.

How SZN Works Through iMessage

SZN works through a single conversation, handling the follow-through and returning when user input is needed.

• Send a request: Text, voice note, photo, or forwarded email.
• SZN acts as a contact: Its name, phone number, and email let it communicate with people and businesses.
• It does the legwork: It can make calls, send emails, find times, and follow up.
• It uses Apple data: Shared Apple Notes, Find My location data, and iCloud files can support tasks.
• It returns for decisions: The agent comes back when your input is required.

Privacy and iMessage Access

Gupta says each customer receives a dedicated model and personal data is not used for training, although these claims have not been independently audited. The bigger open question is how SZN connects to iMessage. Apple has previously blocked unofficial access, including Beeper Mini, while Poke became the first AI agent approved for Messages for Business in June. Whether SZN uses that route remains undisclosed.

What Can SZN Actually Do?

SZN focuses on actions rather than conversation:

Trust becomes important when an agent can spend money. Tinuiti found that 48% of people trust AI to recommend products, but only 20% trust it to make purchases. SZN’s decision-based workflow therefore matters, while pricing remains undisclosed and the preview is currently free.

Why SZN Is Gaining Attention

SZN combines a recognizable founder, an iMessage-first experience, and the broader shift toward AI agents that complete tasks. Edison Research at SSRS found that 52% of U.S. adults use AI platforms weekly, while Gartner predicts that at least 15% of day-to-day work decisions will be made autonomously by agentic AI by 2028, compared with 0% in 2024.

Key reasons for the attention:

• Real contact identity: A name, email, and phone number make it feel like a person.
• No new app initially: It starts as an iMessage conversation.
• Apple integration: Notes, Find My, and iCloud work through one thread.
• Growing competition: Poke, Instinct, and Meta’s Muse on WhatsApp target similar behavior.
• Apple’s response: Third-party access to iMessage remains an important question.

Interest alone does not prove reliability. Gartner estimates that only about 130 of thousands of vendors using the “agentic” label offer genuine agent capabilities. SZN still needs to demonstrate that it can complete tasks consistently and safely.

What an iMessage AI Agent Like SZN Can Actually Do?

An iMessage AI agent like SZN can browse websites, make purchases, communicate with vendors, and follow up on tasks instead of only answering questions. A chatbot generates text inside a conversation, while an agent uses tools, accounts, and contact channels to complete tasks. With its own name, phone number, and email, SZN can act as a go-between while keeping the experience inside one iMessage thread.

1. Browse Websites and Research Options

Instead of opening multiple tabs, users can give SZN a request and let it research options. A Contentsquare survey of 1,300 U.S. consumers found that 19% now choose AI assistants as their primary research tool, while 79% consider accuracy the most important quality in AI-powered shopping.

Accenture found that 74% of consumers are ready to let an AI agent shop for them, 56% would share preferred brands, and 37% of brand-loyal consumers would accept another brand if the agent showed it better matched their needs. SZN’s browsing therefore lets it compare options against user preferences rather than relying only on pre-existing knowledge.

2. Make Purchases and Book Services

Purchasing shows the difference between agents and chatbots, but consumers remain cautious:

NMI also found that 76% of respondents would not let AI complete purchases. Among shoppers willing to allow unapproved spending, Doss found 11% would allow $100 or more in one purchase, while only 6% would allow full autonomy. This suggests users are more comfortable starting with low-risk actions.

Apple’s Messages for Business already supports appointment scheduling and Apple Pay purchase flows. SZN’s model addresses the trust gap by returning to the conversation when a decision is required.

3. Communicate With Vendors

SZN’s phone number and email allow it to contact businesses directly rather than simply telling users what to do. Gartner predicts that agentic AI will autonomously resolve 80% of common customer-service issues by 2029, while 70% of customer-service journeys could begin and end with third-party conversational assistants on mobile devices by 2028.

Instinct is taking a similar approach through Instinct Concierge, which handles phone calls, and its Trusted Person Network, which allows agents to coordinate.

4. Manage Tasks and Follow-Ups

The agent’s value also comes from handling what happens after the initial request. SZN’s product positioning includes emails, calls, follow-ups, scheduling, and checking again later. Poke covers similar tasks through daily planning, calendar management, reminders, and health tracking.

A typical follow-up loop:

• Send the request: Text, voice note, photo, or forwarded email.
• Start the task: Contact the relevant people or websites.
• Track progress: Remember replies and previous interactions.
• Follow up: Check again if there is no response.
• Return for decisions: Ask the user when input is required.

5. Work With Personal Data

SZN is reported to work with shared Apple Notes, Find My location data, and iCloud files, allowing it to use existing context instead of requiring users to re-enter information. Gupta says customers receive a dedicated model and personal data isn’t used for training, although these claims haven’t been independently audited.

Privacy remains a major concern. One U.S. consumer survey found 96% of adults have at least one AI concern, while NMI found 69% don’t trust AI to process payments securely and 84% have never given an AI tool their payment details.

6. Act Beyond Simple Conversations

The distinction becomes clearer when comparing the two:

How SZN Connects to iMessage: What We Know and What’s Unclear

SZN has not publicly explained how it connects to iMessage. Reports confirm that it operates inside a Messages thread with its own phone number, email, and name, but not whether it uses Apple’s official Messages for Business route or an unofficial method. That distinction matters because Apple has previously blocked unofficial iMessage access, most notably with Beeper Mini. Until SZN explains its approach, its long-term reliability remains an open question.

What SZN Has Revealed About iMessage Access

Nikhil Gupta has discussed what SZN does but not its underlying iMessage connection. Bloomberg coverage and SZN’s site describe an agent operating inside an iMessage thread, with a dedicated app planned later.

The key gap is the underlying connection method. Outlets including MacDailyNews and AIBase have highlighted the same uncertainty, while pointing to Poke’s approval as a possible official route.

Why Apple’s iMessage Restrictions Matter

Apple treats iMessage as a core part of its ecosystem and has never released an iMessage app for Android. The audience is also significant: 88% of U.S. teens own an iPhone, according to Piper Sandler, while Counterpoint reported the average iPhone selling price has surpassed $900.

For businesses, this creates platform dependency. Apple controls access and can change its requirements, making unofficial integrations particularly vulnerable.

Key risks include:

• Access can be revoked: Unofficial methods have been blocked within days.
• Official routes still have rules: Approved agents must meet Apple’s requirements.
• Users can be affected: Some Beeper users reported losing iMessage access on their Macs.
• Fallbacks matter: SMS, RCS, or a dedicated app can reduce dependency.

What Beeper Mini Tells Us

Beeper Mini, founded by Pebble creator Eric Migicovsky, reverse-engineered iMessage to bring blue-bubble messaging to Android. Its December 2023 timeline shows how quickly Apple can disrupt unofficial access:

Beeper was later acquired by Automattic. The broader lesson is that unofficial access can work initially but remains vulnerable to rapid blocking and repeated disruption. Beeper even offered its codebase for independent security review, while four lawmakers wrote to the DOJ about Apple’s conduct.

SZN’s Integration: What’s Still Unclear

Several questions remain unanswered:

• Which route does SZN use? Messages for Business or another method?
• How does it obtain its iMessage phone number? This was central to Apple’s action against Beeper Mini.
• How will Apple respond? Observers are watching the product’s October launch.
• How does it access Apple Notes, Find My, and iCloud? The capability is reported, but technical details are private.
• What will it cost? Early access is free and pricing is not announced.

These are open questions rather than criticisms. Current information comes from reports and SZN’s own site, so details may change after launch.

Official vs Unofficial iMessage Access

The official route involves more requirements but provides a legitimate path into Messages. The unofficial route can be faster but carries the fragility Beeper Mini demonstrated. For businesses building an iMessage AI agent, an approved messaging channel plus a fallback is the more defensible architecture. If SZN uses the official path, it would follow this model; otherwise, it carries similar platform risks.

How to Build an iMessage AI Agent Like SZN?

To build an iMessage AI agent like SZN, start with focused use cases, connect through a legitimate iMessage channel, add memory and integrations, then introduce browser control, autonomous follow-ups, and security. The seven building blocks are use cases, interaction, memory, integrations, browser control, autonomous execution, and approval controls.

1. Define Agent’s Core Use Cases

Start with the jobs, not the technology. SZN focuses on email, calls, follow-ups, scheduling, and remembering context; Poke covers planning, calendars, reminders, health tracking, smart-home control, and photo editing; Instinct focuses on travel, reservations, inboxes, and phone calls.

Risk should determine the order. NMI found 63% of consumers would let AI insert discount codes and 58% would let it fill shipping addresses, but 76% would not let it complete purchases. Gartner expects 40%+ of agentic AI projects to be canceled by the end of 2027 because of cost, unclear value, or weak risk controls.

2. Design the iMessage Interaction Layer

For an official iMessage experience, Apple’s Messages for Business requires businesses to work through an Apple-approved Messaging Service Provider (MSP). Apple also expects asynchronous messaging with live agents during business hours. Poke had to demonstrate live support, AI labeling, link previews, and compliance with Apple’s button style.

The official setup is:

• Choose an Apple-approved MSP.
• Register in Apple Business Register.
• Connect and test the provider.
• Submit the experience for review.
• Go Online after approval.

Some vendors advertise direct iMessage APIs from $39/month, but these are vendor claims and carry platform risk. Beeper Mini demonstrates what can happen with unofficial access. A fallback such as SMS, RCS, or Telegram can reduce dependency; Poke already supports these channels.

3. Build Personal Memory and Context

Memory makes the agent feel personal. SZN describes remembering previous conversations and checking again later, while Gupta says each customer receives a dedicated model and personal data isn’t used for training.

Privacy should be designed alongside memory. A U.S. consumer survey found 96% of adults have at least one AI concern. Store only necessary information, separate users’ data, and give users controls to inspect and delete memories.

4. Connect Tools, APIs, and Accounts

An agent needs access to the systems behind its tasks. MCP has 97 million monthly SDK downloads, and Anthropic donated it to the Agentic AI Foundation under the Linux Foundation in December 2025. Poke uses API and MCP integrations on hosted infrastructure, according to Chorus.

For a personal agent, start with email, calendar, messaging, payments, and user-shared data.

• Use scoped permissions: Request only what each task requires.
• Separate read/write access: Reading a calendar is safer than sending messages or moving money.
• Log actions: Record what the agent did and why.
• Expect failures: Verify tool results before continuing.

OWASP identifies excessive agency as a distinct LLM application risk.

5. Add Browser and Computer Control

Some services have no APIs, so the agent needs browser or computer control. SZN is reported to browse websites and make purchases, while Instinct operates phones and computers. Poke’s API/MCP approach is narrower because it cannot log into arbitrary websites.

Browser access also increases security risks. OWASP ranks prompt injection as the number-one LLM application risk. Anthropic tested Claude for Chrome with 123 adversarial cases and found a 23.6% attack success rate, reduced to 11.2% with mitigations. Anthropic does not recommend autonomous mode.

Build Rule: Treat everything the agent reads on the web as untrusted, and never allow page content to authorize a purchase, message, or data share.

6. Enable Autonomous Tasks and Follow-Ups

Autonomous execution is what separates an agent from a chatbot. The system needs to keep tasks open, wait for external events, retry failures, and return when user input is required. SZN follows this model, while Poke is described as proactive. Gartner predicts 15% of day-to-day work decisions will be made autonomously by agentic AI by 2028, compared with 0% in 2024.

7. Secure Actions With Approval Controls

Approval controls determine what the agent can do automatically. Divide actions into risk tiers:

OWASP recommends human approval for high-risk actions and least-privilege access. Claude for Chrome follows the same principle before publishing, purchasing, or sharing personal data.

How Much Does it Cost to Build an iMessage AI Agent Like SZN?

Building an iMessage AI agent like SZN typically costs 15,000–45,000 for an MVP, 50,000–150,000 for an advanced agent, and 150,000–500,000+ for an enterprise system, plus ongoing costs. The biggest cost drivers are integrations, autonomy, security, and whether you use Apple’s official Messages for Business channel.

MVP iMessage AI Agent Cost

Estimated cost: 15,000–45,000, built in 8–12 weeks.

An MVP should focus on one job. Empat estimates AI proofs of concept from $15,000 and AI MVPs from $30,000 over 8–12 weeks, while SoluLab estimates 8,000–45,000 for a single-purpose agent. Poke also started with narrower tasks such as planning, reminders, and calendar management.

A typical MVP includes:

• One end-to-end use case: Scheduling or vendor follow-ups.
• One text channel: SMS/RCS first, with iMessage through an approved provider.
• 2–3 integrations: Calendar, email, and one booking tool.
• Basic memory: Thread context and saved preferences.
• Approval prompts: Required before sending, spending, or sharing.
• Excluded: Browser control, multi-day autonomy, and enterprise compliance.

Advanced Agent Development Cost

Estimated cost: 50,000–150,000, built in 3–6 months.

This tier resembles SZN or Poke, with long-term memory, multiple integrations, browser control, and scheduled follow-ups. Empat estimates custom AI agents at 30,000–150,000, Silent Infotech lists 75,000–150,000+, and Creole Studios estimates 100,000–250,000+ for autonomous agents.

Enterprise-Grade Agent Cost

Estimated cost: 150,000–500,000+, built in 6–12 months.

Enterprise systems add multi-agent coordination, compliance, audit logs, SSO, stricter data controls, and potentially dedicated models. Empat estimates enterprise AI platforms at 100,000–500,000+, while Creole Studios puts enterprise multi-agent systems at 300,000–500,000+.

Security can significantly increase costs if added late. Codewave reports that retrofitting security and regulatory controls can double implementation costs. Apple’s Messages for Business review also adds time before launch.

Key Factors Affecting Development Cost

Two agents with similar features can have very different budgets. The main drivers are integrations, autonomy, multi-agent architecture, security, data readiness, and channel choice.

Greenice analyzed 500+ AI agent projects on Upwork and concluded businesses often underestimate custom agent costs by 3–10×. A practical approach is to tightly scope the MVP and expand based on real usage.

Ongoing AI Agent Running Costs

Development is only the initial investment. Ampcome recommends budgeting 15%–20% of the build cost annually for maintenance, while Creole Studios estimates 500–1,500+/month for small MVPs and $10,000+/month for enterprise systems. SoluLab estimates 150–1,500/month for a single-purpose MVP, while managed vector databases can cost 200–800/month.

Anthropic lists mid-tier Sonnet models at $3/M input tokens and $15/M output tokens, while Haiku 4.5 costs $1 and $5.

Worked Example

A request with 2,200 input tokens and 400 output tokens costs about $0.0126 on a Sonnet-tier model and $0.0042 on Haiku 4.5. At 100,000 requests/month, that becomes roughly $1,260 vs. $420. Agent tasks usually involve multiple model calls.

Prompt caching can reduce cached input costs by up to 90%, while batch processing can halve input and output prices. Routing simple tasks to cheaper models can further reduce costs.

iMessage Integration and API Costs

Apple requires businesses to use an approved Messaging Service Provider (MSP) for Messages for Business, but Apple does not publish a standard price list. Providers set their own rates, so businesses need quotes. Poke’s co-founder has said it pays per user, while SZN has not announced pricing.

At $0.0083/message, 100,000 messages cost about $830 before carrier fees.

For budgeting:

• Request quotes from multiple approved providers.
• Price a fallback messaging channel.
• Estimate messages per user and total users.
• Add expected voice minutes.
• Keep a buffer for provider changes and Apple’s review timeline.

Apple’s Approval Requirements for AI Agents (Checklist)

To get an AI agent approved on Apple’s iMessage business channel, you work through an Apple-approved Messaging Service Provider, verify your business in Apple Business Register, clearly label the agent as AI, provide live human support, follow Apple’s messaging guidelines, and pass the Experience Review. 

Poke became the first AI agent approved on Messages for Business after showing it could offer live support and clearly identify its AI agent. Apple has no separate AI-agent rulebook, so this checklist combines its general Messages for Business policies, provider documentation, and Poke’s approval precedent.

Approval checklist at a glance

• Choose an Apple-approved Messaging Service Provider
• Register and verify your business in Apple Business Register
• Clearly identify the agent as AI at the start of conversations
• Provide live human support
• Follow Apple’s messaging and brand guidelines
• Test every path on a real device
• Pass the Experience Review and submit commercial details

1. Approved Messaging Provider

Apple requires businesses to work through an approved Messaging Service Provider. The provider supplies the messaging platform, Apple connector, live-agent console, and bot-to-human handoff tools. Providers must support asynchronous conversations, intent-based routing, and live-agent handoff with conversation history.

Poke followed this route and reportedly pays its provider per user. Apple itself charges the business nothing; providers set their own pricing.

• Is the provider on Apple’s approved list?
• Does it support bot-to-person handoff with conversation history?
• Who handles the Experience Review?
• What are the setup, message, and exit fees?
• Does it support AI-agent workflows?

2. Complete Apple Business Verification

Verification happens through Apple Business Register. You need an Apple ID, administrator access, and an approved provider. Apple’s requirements also expect live-agent support during business hours and service comparable to phone support.

3. Clearly Identify the AI Agent

The agent must identify itself as AI at the start of the conversation. Poke’s approval also depended on clearly identifying its agent as AI. This is especially important for products like SZN that give the agent a name, phone number, and email address. Your product naming must also remain consistent across the help center, buttons, and scripts.

• The first message identifies the AI agent
• The AI label appears after handoffs or resets
• Live agents are introduced when they join
• Product naming is consistent
• Apple’s product names are used correctly

4. Provide Live Human Support

Apple requires access to a live agent during regular business hours; a bot-only experience isn’t sufficient. Poke had to demonstrate that it could provide live support when needed.

• Live agent access: Customers must be able to request a person.
• Escalation: Unresolved requests should move smoothly to a human.
• Availability messaging: Outside business hours, explain when help is available.
• Agent introduction: Identify live agents after transfer.
• Context transfer: Pass the conversation history to the human agent.

5. Meet Apple’s Messaging Guidelines

Messages for Business is designed around customer-initiated conversations. Users can start conversations through a Message Us button, QR code, shipping email, Apple Maps, or Message Suggest. Poke also had to use link previews and follow Apple’s style guide.

6. Test the Agent Before Review

Apple reviews the full customer journey, including entry points, automation, resolution, and live-agent handoff. Test everything on a real iPhone. Providers may also require an iPhone screen recording showing rich features, live-agent interaction, and out-of-hours messaging.

7. Complete Apple’s Final Review

The final stage includes the Experience Review and commercial account review. Apple typically requires one to three Experience Review rounds, while commercial review usually takes one to three business days.

• Experience Review: Apple checks the test environment end to end.
• Commercial details: Submit your commercial account information.
• Commercial review: Usually takes one to three business days.
• Go Online: Request final approval with live agents ready.
• Re-review: Major new flows or features may require another review.

Budget additional time for approval. Poke’s approval reportedly took several months, and it had relayed about 100 million messages across supported platforms before launching on Messages for Business. Apple can also update its requirements, so verify the current rules in Apple Business Register.

Competitive Landscape: SZN vs Poke vs Instinct 

SZN, Poke and Instinct all place personal AI agents inside messaging, but their strengths differ. SZN focuses on Apple integration, Poke on proactive assistance and Apple approval, and Instinct on task execution, including phone calls. None has a settled lead. Their potential advantages are more likely to come from trust, channel access, distribution and reliability than the underlying AI model. Meta’s Muse on WhatsApp adds another major contender with much larger distribution.

SZN: iMessage-First Personal AI

SZN was created by former Siri and Apple Intelligence engineer Nikhil Gupta through Nomadic Futures, Inc. It starts in iMessage without a separate app, although a dedicated app is planned. Users can send text, voice notes, photos or forwarded emails to an agent with its own name, phone number and email. 

It reportedly works with Apple Notes, Find My and iCloud. Gupta says each customer gets a dedicated model and personal data isn’t used for training. Early access is free, with a wider launch planned for October.

What SZN has going for it

• Apple-ecosystem depth: Notes, Find My and iCloud work through one thread.
• Personal identity: Its own number and email enable external interactions.
• Privacy pitch: Dedicated models and no training on personal data.

What is still open

• iMessage access: Its connection method hasn’t been disclosed.
• Pricing: Not announced.
• Reliability: Still unproven at wider scale.

Poke: Multi-Channel AI Assistant

Poke, built by The Interaction Company of California, runs on Apple Messages, SMS/RCS and Telegram, with WhatsApp available in some markets. It focuses on proactive help, including schedule conflicts and flight changes. Recipes supports automations, while Poke Kitchen helps users create them. 

A review lists more than 40 service connections, including Gmail, Google Calendar, Notion, GitHub, Linear and Philips Hue. Poke Human can hand tasks such as bookings and calls to a person.

Cognition’s interest also reflects the importance of personality. Von Hagen told TechCrunch that people prefer colleagues with personality over robotic systems.

Instinct: iMessage, WhatsApp, and SMS

Instinct is built by Spear Street Technology and led by Noah Shinn. It launched as an invite-only service in August across iMessage, WhatsApp and SMS, without a mobile app. It uses its own phone number and computer to operate websites like a person. TechCrunch reports that it handles reservations, purchases, bill payments, cancellations and grocery orders, with more than half of transactions reportedly travel-related.

• August: Invite-only launch and a $250 million Series B at a $2.5 billion valuation.
• September 16: Instinct Concierge began making phone calls, including restaurant bookings and dentist cancellation-list requests. It also added personal email addresses and the Trusted Person Network.
• September 28: A $1 billion Series C at a $10 billion valuation.
• Late September: Instinct Selections launched for product recommendations and drew complaints from some users.

TechCrunch noted that Instinct had not disclosed user count, revenue, pricing or a public launch date, while privacy criticism led it to revise its terms.

How These Agents Differ

SZN focuses on deep Apple integration, Poke on proactive assistance and multi-channel access, while Instinct emphasizes autonomous task execution through messaging, browsing, and phone calls. Their differences mainly come down to channels, integrations, autonomy, availability, and trust. 

Where the Competitive Moat Lies

The underlying models are increasingly interchangeable, so differentiation sits elsewhere:

• Channel access. Poke was described as the only AI agent natively supported by Apple Messages when Cognition announced its acquisition. Other agents need an approved route or face risks like Beeper Mini.
• Trust and permissions. 69% of consumers don’t trust AI to process payments securely, according to NMI, while a Doss survey found only 6% would allow an agent to make purchases entirely on its own. Meta’s Muse uses Muse Secure VM, Sentinel permissions and Stripe Link one-time cards.
• Distribution. Muse launched on September 8 for U.S. adults across iOS, Android, web and WhatsApp, then added phone calling in mid-September. WhatsApp reaches more than 2 billion people.
• Reliability. One review roundup scored Poke 2.9 out of 5 for reliability, citing issues with time zones, dates and slow integrations. Gartner estimates only about 130 of thousands of vendors calling themselves agentic offer genuine agent capabilities.
• Personality and memory. Cognition’s acquisition of Poke highlights how conversational personality and user memory can also differentiate an agent.

Develop an iMessage AI Agent With IdeaUsher

Build a custom iMessage AI agent with IdeaUsher, backed by 500,000+ hours of coding experience and a team of ex-MAANG and FAANG developers. We help businesses turn conversational AI concepts into secure, scalable agents that can understand context, access tools, and execute real-world tasks. From MVP development to enterprise-grade deployments, our team supports the complete development lifecycle.

Custom Agent Architecture and Development

Design and develop an agent architecture tailored to your use cases, autonomy requirements, workflows, and scalability needs. We can build modular architectures that support multi-step reasoning, task orchestration, memory, and controlled agent actions. 

iMessage Infrastructure Integration

Build the messaging layer and integrate your agent with the appropriate iMessage infrastructure, including authentication, messaging workflows, handoffs, and platform requirements. We also help structure the integration around Apple’s messaging requirements and scalable communication workflows.

LLM, Memory, and RAG Implementation

Integrate LLMs with persistent memory and RAG to help your agent retain relevant context, personalize responses, and retrieve information when needed. This enables more contextual conversations while helping the agent work with user-specific information and knowledge sources. 

Tool and API Integration

Connect calendars, email, payments, CRMs, databases, third-party APIs, and other tools so your agent can move beyond conversations and complete real-world tasks. Our developers can create secure tool workflows with permissions, API orchestration, and approval controls for consequential actions. 

Conclusion

Building an iMessage AI agent like SZN with its own mobile number requires more than connecting an LLM to a messaging interface. It combines AI agents, memory, tool integrations, autonomous workflows, and secure messaging infrastructure to create an assistant that can communicate and act on the user’s behalf. With the right architecture and platform strategy, businesses can turn this concept into a scalable personal AI agent that handles real-world tasks through a familiar messaging experience. 

FAQs