This is the fourth post in my AI Agent Series, where I walk through each agent I run operationally in my business and show you exactly how it works. This one handles remote task dispatch via iMessage.
The previous posts covered the morning email agent, the VOC ad copy agent, and the Klaviyo flow builder. Each one removes a specific execution bottleneck. This one removes the bottleneck that connects them all: the requirement to be at a desk before anything can happen.
The Before: Everything Required a Laptop
Every creative impulse, every client task, every operational action had the same tax attached to it: get to a laptop, open the right tools, load the right context, and start from zero. The idea you had in a cab at 8am sat in a note somewhere until you could act on it. The CRO audit you wanted to run for a prospect before a call took 45 minutes to set up properly. The blog post you needed to publish required cloning a repo, writing code, and deploying manually.
That friction is invisible until you remove it. Once you do, you realise how much creative and operational momentum gets lost between the moment an idea arrives and the moment you can actually act on it. Most operators have trained themselves to accept that gap as normal. It is not normal. It is a systems problem with a systems solution.
What the Hey Clawd Agent Does
Hey Clawd is a custom iMessage command interface built on Claude's agent SDK. It runs on a schedule, polling my iMessage thread for new messages. When it finds one starting with "Hey Clawd", it parses the request, identifies the task type, routes it to the relevant agent skill, executes the full workflow, and sends the result back as a reply in the same thread.
The interface is a single iMessage conversation. The capability behind it is a full business operating system.
Input
Hey Clawd — write a blog post on TikTok Shop affiliate seeding for supplement brands
Output
Researches the topic, writes a 2,000-word post in my voice, deploys it to the live site, and replies with the URL. This post was written and published while I was away from my desk.
Input
Hey Clawd — run a CRO audit on [brand URL] before my call tomorrow
Output
Crawls the site, scores it across 9 CRO categories, builds a live HTML report with revenue uplift forecasts, and replies with the link. Ready for the call.
Input
Hey Clawd — pull VOC from [brand] reviews and write 3 Meta ad angles
Output
Mines reviews across Google, Amazon, and Trustpilot, extracts the top objections and desire drivers, writes three direct-response ad angles with headlines and body copy, and saves them to the client folder.
Input
Hey Clawd — draft a win-back sequence for [brand] in Klaviyo
Output
Reads the brand's existing email context, drafts a 3-email win-back sequence with subject lines and copy, and flags it for review before pushing to Klaviyo.
Input
Hey Clawd — research [prospect company] and build a growth proposal
Output
Scrapes the prospect's site and social presence, identifies their top growth levers, builds a branded proposal deck, and saves it to the client folder with a Google Drive link.
The Brand Memory and Context Layer
What separates Hey Clawd from a generic AI assistant is what it knows before the message arrives. The agent loads a set of context files at the start of every session: who I am, what the business does, which clients I am working with, what has already been published, what the Purposeful Profits design system looks like, what my writing voice sounds like, and what integrations are available. It does not start from zero on every task. It starts from where I am.
This is the layer that makes it feel like a trained team member rather than a tool. When I text a blog post request, it already knows the existing slugs and will not create a duplicate. When I ask for a client proposal, it knows that client's retainer size and the commercial model we use. When I ask for ad copy, it loads the brand's voice, banned phrases, and current messaging pillars. The context is persistent. The agent compounds over time as I add to it.
What the Output Actually Looks Like
The agent always replies with either a URL, a file path, or a structured summary — never just a block of text with no place to go. A deployed blog post comes back with the live URL and a ready-to-post LinkedIn draft. A CRO audit comes back with a hosted report link. An email draft comes back with the subject lines, full copy, and a confirmation of where it was saved. A client proposal comes back with the Google Drive link and a note on what was included.
The agent does not deliver drafts that live in a chat window. Every output lands in the right place: a live URL, a file on Drive, a sequence in Klaviyo, a message in iMessage. The work is done, not described.
One thing the system does well is confirmation. For anything that touches a live customer-facing system, the agent flags before it acts. Blog posts get deployed and the URL is sent for review before the database email goes out. Email sequences are drafted and saved for human approval before any flow is activated in Klaviyo. The agent operates with real autonomy on research, drafting, and building. It defers on publishing and sending. That balance matters in practice.
Why iMessage Specifically
There are plenty of ways to interact with an AI agent. A web dashboard, a Slack bot, a dedicated app. I chose iMessage because it is where my phone already lives. Every DTC operator I know checks iMessage dozens of times a day. It requires no new habit, no new app, no new context switch. The command interface lives in the same thread as messages from clients, friends, and family. It is frictionless by design.
The market has validated this bet. In June 2026, Apple approved Poke as the first official AI agent on its Messages for Business platform. The startup has already relayed over 100 million messages and raised at a $300 million valuation. The operators who recognise that the phone, not the laptop, is the natural control surface for an AI-assisted business will have a compounding advantage over those still requiring a desk to get anything done.
The Compounding Effect on Operator Leverage
The individual time savings are real but they are not the main point. The main point is what happens to your creative and strategic throughput when the friction between idea and execution drops from hours to seconds.
Before this system existed, a good idea at 7am had maybe a 30% chance of becoming a real output by the end of the day. Between meetings, context switching, and the accumulated mental load of everything else on the plate, most ideas died between thought and execution. After the system, the conversion rate on ideas is close to 100%, because the execution cost is a 30-second text message.
In a business where content, client work, and growth experiments compound over time, that difference in throughput is the difference between an average year and an exceptional one. You are not saving time in isolation. You are multiplying the leverage of every idea you have.
Inside the system
How we build this for brands
Hey Clawd is one node in a broader agent stack built on Claude. The other nodes — the VOC engine, the Klaviyo flow builder, the CRO audit system, the morning email triage — are what give the iMessage interface its capability. The command layer is simple. The power comes from the skills and integrations wired behind it. Each skill is trained on the specific outputs a DTC brand needs: email sequences in the right voice, ad copy built from real customer language, audit reports that name the revenue impact rather than just listing issues.
We run a version of this across portfolio brands, connected to their Shopify data, their Klaviyo accounts, and their brand context files. The iMessage interface is the entry point an operator sees. The full system is what executes underneath it. Part of this runs live for portfolio brands today. The complete build, configured around a specific brand's data and growth targets, is what we deploy when we take a brand on.
Build Your Agent Team
Get your own AI agent team working on your brand
Caner can help you build your own AI agent teams and unlock growth for your brand. This is not a template or a SaaS tool. It is a custom system designed around your data, your workflows, and your growth targets.
Book A DemoFrequently asked questions
Can I build an iMessage AI agent like Hey Clawd myself?
Technically yes, if you have experience with Claude's API, agent orchestration, and Mac automation tools. The system uses Claude's agent SDK, a scheduling layer, AppleScript for iMessage interaction, and a suite of pre-built skills for specific tasks. In practice, building this from scratch takes weeks of engineering and configuration. The more efficient route for most DTC operators is to work with someone who has already built and battle-tested the system, so you get the output rather than spending time on the infrastructure.
How long does it take to set up an AI agent like this?
The core iMessage interface and scheduling layer takes two to three days to configure. The more significant time investment is training the agent on your business context: your brand voice, your client portfolio, your existing content, your integrations with Shopify and Klaviyo. A well-configured system that genuinely knows your business takes one to two weeks to set up properly. After that, the ongoing time cost drops to near zero.
What tasks can the Hey Clawd agent actually handle?
Hey Clawd can handle any task that has been built as a skill or workflow in the agent stack. Current capabilities include deploying new blog posts to the live site, running CRO audits on any URL, drafting and scheduling email campaigns in Klaviyo, mining customer reviews for ad copy, generating ad creative briefs and image prompts, pulling performance data from connected platforms, researching new prospects, and drafting client proposals. New capabilities are added as new skills are built, so the system grows with the business.
Is this just a ChatGPT wrapper that sends iMessages?
No. The distinction matters. A ChatGPT wrapper generates text responses. Hey Clawd takes action: it clones a GitHub repo, writes and deploys code, calls external APIs, sends emails, reads and updates files on connected drives, and returns URLs, reports, and deliverables, not just text. The agent has access to a full toolkit and executes multi-step workflows autonomously. The iMessage interface is simply the command layer. The actual work happens in an agent system with real integrations.
What happens if the agent makes a mistake or misinterprets a task?
The agent is configured with a review layer for high-stakes actions. Blog posts are deployed and a notification is sent with the live URL before any email goes to the database. Proposals are drafted and flagged for approval before sending. The agent confirms before taking irreversible actions. For most tasks, the output comes back as a draft or a link, which the operator reviews before it touches customers. The system is designed to generate fast drafts for human review, not to act unilaterally on live customer-facing systems.
About the author
Caner Veli built Liquiproof to global distribution across 3,000+ retailers, then exited. He now runs Purposeful Profits using a combination of operator strategy and AI-powered systems he has built and uses daily, having 10x'd monthly revenue in his own business in the last 90 days.