The Most Expensive Message Is the One You Never Send
Every cross-border seller I know is carrying a small pile of unread messages they intend to answer “later” — a supplier waiting on a PO confirmation, an Amazon buyer with a damaged unit, an influencer who went quiet after the last DM and now writes for a competitor. Later is expensive. That’s why the Product Hunt launch that caught my eye this week wasn’t the promoted AI site builder promising to design and publish professional pages with a prompt. It was Blueberry, a Mac menu bar app from Rishi Narain that finds the iMessage threads you’ve ignored and auto-drafts a reply in your voice. It’s a tiny product, and it doesn’t pretend to replace your messaging app. But it embeds a working answer to the most overlooked operations problem in cross-border trade: activation energy.
What Blueberry Actually Solves
Rishi frames the problem in personal terms: “I see a text while I’m busy, think I’ll respond to this later, and completely forget about it.” The buried message then disappears under spam and group noise. He says this has cost him work opportunities, friendships, and dates. For a cross-border seller, the same dynamic runs through the supply chain. A factory manager sends a shipping update; you read it at 2 AM and don’t reply; by the time you come back, your restock lead time has slipped. A buyer asks whether a gift box can arrive before Saturday; no answer for two days; they leave a negative review. The cost of a forgotten follow-up is rarely a line item in a P&L, but it is often larger than the ACoS spread you just spent a month optimizing.
Blueberry’s mechanics are intentionally narrow. It scans iMessage for threads that need a reply, auto-drafts a response in your voice, and lets you view, edit, or pick an alternative variation before sending it in one click. It works in individual and group messages. It includes an “unhinged mode” and a “flirt mode” for consumer fun. It never sends automatically — you always hit send, and there is a five-second unsend window in case you change your mind.
The maker’s clearest statement is the one operators should care about: “we don’t want to replace your iMessage app or automate your texting.” This is the opposite of most AI customer-service tools, which aim to close tickets without a human. Blueberry’s goal is to remove “the small bit of friction that turns ‘I’ll answer this later’ into three months of accidental ghosting.” That friction is what kills cross-border relationships: time zones, language barriers, and the thousand tabs of an operator’s day.
Why Amazon sellers should care more than Shopify ones
Amazon sellers have strict seller metrics around buyer-seller messaging. Late replies can hurt your account health, and fully automated replies that sound canned are both easy for buyers to ignore and risky under Amazon’s communication policies. A draft-and-approve model is a better fit. AI proposes, a human approves, and the send button stays yours. Shopify merchants, by contrast, own the chat experience and can safely deploy Gorgias with broad automation because their customers expect an instant brand bot. On Amazon Seller Central, the buyer is Amazon’s customer first; your reply is a customer-service event, not a branded conversation. The discipline Blueberry encodes — no message goes out until you click send — is more valuable to an FBA operator than to a DTC Shopify brand.
How It Differs from the AI Tools You’re Already Using
The obvious comparison is ChatGPT. You can paste a garbled supplier message into a prompt and ask for a reply. But that workflow has a serious flaw: the model doesn’t see the other twelve messages from the same person, doesn’t know which thread is urgent, and has no idea whether your tone should be formal with a factory manager or casual with a freight forwarder. Blueberry reads recent thread context to generate drafts. That is the difference between “an AI that can write” and “an AI that knows what to write about.”
The next comparison is email-side AI assistants like Superhuman. They are good at triage and AI compose, but they live in email. Cross-border e-commerce doesn’t primarily run on email. It runs on WhatsApp, WeChat, and Messenger. Blueberry is iMessage-only now, but when a commenter named Nika says her Facebook messages have been unanswered since 2021, Rishi’s reply is immediate: “we’re bringing messenger in next.” That “next” is exactly where cross-border operators already live. The best AI messaging assistant will not be the one with the most sophisticated language model. It will be the one that connects to the channel where your money is stuck.
Finally, compare Blueberry to rule-based text expanders or snippet managers. Those are static. They can’t judge which message in a group thread is the one that needs a reply. Blueberry’s “voice” is an attempt to solve tone, which is why Igor Gurovich, a founder building voice AI at Callie Care, asks whether tone shifts with relationship closeness and how many corrections it takes before drafts actually feel like you. The launch post doesn’t answer that. But the question itself is a good design spec: a draft that sounds like generic AI will be edited, and if the operator has to edit it, the app has failed at removing friction.
Where the math breaks
The launch includes a code from Dean Allen for “30 days of free pro access,” but the actual price is not disclosed. That alone is a red flag for operators who need predictability. More importantly, the product requires Full Disk Access to read iMessages and Automation permission to send replies. Recent thread context goes to the cloud to generate drafts, with models hosted on AWS Bedrock or Google Vertex. The maker promises model providers cannot see or train on your data. That may be true, but it is a trust promise, not an audit. If your phone receives payment details, supplier price lists, or legal claims through iMessage, “trust us with Full Disk Access” is a hard no in a procurement review. Also, Mac-only and iMessage-only means most teams on Windows, Android, and WeChat can’t use it. The math only works if your personal iMessage is both your relationship layer and your business inbox. Few cross-border operators have that luxury.
What Cross-Border Sellers Should Steal From Blueberry
Don’t copy the app. Copy the interaction design. The “draft, approve, send, undo” loop is the future of AI in operations. Here is how to apply it this week.
First, put a human between your AI and your customers. If you use Gorgias or an email platform with AI-suggested replies, configure it so no message goes out without a click. The small cost in response time is worth the control over refunds, policy promises, and tone.
Second, build a voice file. Write ten sentences describing how you talk to suppliers, how you talk to VIP customers, and how you talk to strangers. Feed that into the AI tool you use for drafting. Blueberry tries to learn your voice from context; you can do the same manually with a prompt. The Igor Gurovich question about relationship-based tone should be your calibration checklist. A supplier wants a no-BS first-name reply. A buyer in the EU wants clarity and warmth. A TikTok influencer wants hype. One voice does not scale.
Third, add a cooling-off period to any message that contains emotion. Blueberry’s five-second unsend is a small version of “write the angry review response at night, send it in the morning.” If your operations stack doesn’t support that, build a draft folder in your CRM before you let AI touch outbound messages.
Fourth, treat group chats as high-risk zones. Group messages are where the cost of AI hallucination is highest: a joke taken literally, a price error quoted to three suppliers at once. Blueberry supports group messages, but that support should make you more careful, not less.
The Ghosting Audit
Here is a twenty-minute exercise for any founder or operations lead. Open WhatsApp, WeChat, and the buyer-seller inbox in Amazon Seller Central. Count the threads with messages older than twenty-four hours that you intended to answer. Estimate the potential order value in each one. Most operators will find at least five threads worth more than a thousand dollars. Then take one and ask ChatGPT to draft a reply in your voice, with context: “I run an e-commerce brand, this supplier has been waiting on a PO change, I don’t want to sound robotic, I want to apologize for the delay.” Edit it, send it, and schedule the same audit for next Monday. That is Blueberry’s entire value proposition, minus the Full Disk Access.
Where My Judgment Says It Falls Short
For all its charm, Blueberry is not a tool I would build a cross-border operation around. First, it is platform-limited. A Mac menu bar app that only reads iMessage misses the reality of global trade: factory managers use WeChat, freight forwarders use WhatsApp, and Amazon customers use email through Seller Central. “Messenger in next” is a step, but it is not enough to make Blueberry the unified inbox a cross-border seller needs.
Second, the privacy story is not as simple as the maker’s bullet points suggest. Full Disk Access is a powerful permission. Even if the app never sends automatically and model providers cannot train on the data, a cloud call with recent thread context is still a data transfer. For a solo founder whose phone contains startup secrets, that is manageable. For a company with finance conversations, supplier contracts, and customer PII, it is a compliance review. You should never approve a tool like this without asking who can access the logs, where they are hosted, and what happens on a breach.
Third, the “voice” claim is unproven. The launch answers a lot of questions, but not the one that matters: how does the app know your voice early on? Igor Gurovich asks exactly this: does tone shift based on relationship closeness, and how many corrections are needed? The absence of a number is typical for a first launch, but it means the central promise is still a demo, not a system. If it takes more than a few corrections to feel like you, the app is a novelty with iMessage access.
Fourth — and this is where the Framer AI Agents card in the same feed is instructive — Blueberry’s restraint is the opposite of the “design and publish professional sites with AI” pitch. I have no issue with AI website builders; they have a place in rapid testing. But for cross-border sellers, a website is a storefront that must handle international payments, VAT, shipping rates, and return logic. An AI agent that publishes a professional-looking site without a human reviewing tax and logistics settings is a liability. Blueberry’s “you always hit send” is a better default for commerce than “just publish it.” The same sentence applies to automated advertising, automated repricing, and automated review requests: the more irreversible the action, the more human-in-the-loop you need.
Fifth, the “5-second unsend” is more psychological than operational. Once a message is sent, the damage is done; the recipient saw it on their watch. In iMessage, there is an undo-send window, but it is not a magic eraser. For an Amazon seller, you cannot unsend a reply that admits a policy violation. The real lesson is to prevent bad messages from being sent in the first place — which means editing the draft before sending, not relying on an emergency undo.
What I’d Watch / Test Next
I’m going to watch whether Blueberry ships a Messenger integration and, more importantly, whether it expands beyond iMessage to WhatsApp and SMS. The moment it does, it becomes interesting on a team level, not just a personal level. In the meantime, I’d test it only on a secondary Mac with a dedicated iMessage account, not on your primary phone. And I’d use the concept, not the app, as the template for your AI stack: drafts in your voice, a human on send, and an undo gesture for the moments when the draft is wrong. This week, run the ghosting audit. Then set a recurring rule: no AI-generated customer or supplier message goes out without your own name on it. The brands that do that will win the response-time game without losing the relationship game. That applies whether you’re selling on Shopify, Amazon Seller Central, or the next channel that makes you re-learn everything.
