We almost built the wrong thing
Three thinking tools that stopped us from building a shiny feature nobody needed.
Last week we sat down to plan the next feature for our AI assistant system. The feature? Instant messaging. Text your AI from your phone, wherever you are.
It is amazing. It's one of those features that makes people's eyes light up in demos. "Wait, I can just... message my AI? From my phone? While I'm walking the dog?"
And that excitement is exactly the trap.
Because "text your AI from anywhere" is a solution. And we hadn't properly defined the problem.
The street performer test
Every time we plan something new, the same question comes up first: What problem are we actually solving?
Not "what would be cool." Not "what would make a great demo." What specific, recurring pain does this fix for someone running a coaching business?
The underlying technology for instant messaging is genuinely interesting. There are protocols and APIs and architectural decisions that matter a great deal if you're the one building it. And none of that matters to you. What matters is the thinking process that happens before you write a single line of code.
So instead of diving into the build, we used three metaphors to pressure-test the idea. Three thinking tools that forced us to get honest about what we were actually building, and for whom.
The executive assistant, not the help desk
This was the first metaphor, and it changed everything.
Think about the difference between a help desk and an executive assistant. A help desk waits for you to show up with a problem. You walk in, take a number, explain what's wrong, and hope someone can fix it.
An executive assistant? They come to you. They know your schedule, your priorities, your patterns. They don't wait for you to remember something — they surface it before you forget.
So the question became: Are we building a help desk with a chat interface? Or are we building an executive assistant that can reach people where they are?
That reframe alone eliminated half our feature list.
The shame problem nobody talks about
I need to tell you something personal, because it's the reason this section exists.
I have executive function challenges. The clinical term is executive function differences — and if you're a high-performing entrepreneur, there's a decent chance you know exactly what I'm talking about from the inside. The brilliant strategist who can see five moves ahead and can't remember to reply to the email they opened twenty minutes ago. That's me. It's probably a few of you reading this, too.
I've tried hiring executive assistants. Real, human, well-intentioned ones. And it made things worse.
The EA was good at their job. That was the problem.
A competent EA proactively checks in. "Hey, you haven't responded to this." "Just following up on that thing from Tuesday." "Did you get a chance to look at the proposal?"
And every single one of those well-meaning nudges landed like a small shame grenade. I already knew I'd dropped the ball. I always knew. Having another human being point it out didn't create action — it created avoidance. I stopped opening those messages. I started dreading the check-ins. The relationship deteriorated. The EA got frustrated. I felt worse.
I've watched this same pattern play out with dozens of founder-coaches I work with. The founder fires the EA (or the EA leaves), and the takeaway is always "I'm just not an EA person." As if it's a personality deficiency rather than a fundamental mismatch between the support system and the way their brain works.
Why the AI version might actually work
The reason an AI messaging assistant might succeed where human EAs failed isn't the technology. It's the psychology.
An AI doesn't judge. It doesn't get frustrated when you ignore it three times. It doesn't sigh. It doesn't update its LinkedIn profile because you're a difficult boss. It has infinite patience.
For founders with executive function challenges, that changes the entire dynamic. The AI can be proactive — surfacing what matters, nudging when needed — without triggering the shame cycle that made human support fail.
You can ignore a message from your AI at 3pm and pick it up at 11pm and it greets you exactly the same way. No guilt. No "as per my last message." No emotional charge.
The technology is incidental. The psychology is the whole game.
The walkie-talkie, not the control room
The second metaphor shaped how we thought about the interface.
A control room is where you do strategy. Big screens, full context, every tool at your fingertips. That's your desk. Your laptop. Your dedicated AI session where you can see everything and think deeply.
A walkie-talkie is what you carry in the field. Quick check-ins. Status updates. "Send backup to sector 7." You're not doing strategy on a walkie-talkie. You're executing.
This told us something crucial: the messaging interface works alongside the full AI workspace. Different tool, different context, different purpose.
Strategy happens at the desk. Execution happens in the field. Both are valid. And they need different interfaces.
The concierge with multiple entrances
The third metaphor was about access.
Think about a high-end hotel. The concierge can book you a restaurant, arrange transport, get theatre tickets, solve problems. Same capability. Same person.
Some guests walk through the lobby and approach the front desk. Others call from their room. Others send a text message. Different entrances, same service.
In the AI assistant world, some people are comfortable in the "back office" — the terminal, the full workspace, the power tools. They love it there. That's their lobby.
Other people would never walk into that lobby. It's intimidating. Foreign. They'd feel out of place.
Give them a messaging interface — a front desk they're already comfortable with — and suddenly the same capabilities become accessible. Not because you built something new. Because you opened a door they were willing to walk through.
This matters enormously for coaching businesses building with AI. Your team members, your clients, your partners — they all have different comfort levels with technology. The underlying capability doesn't change. The entrance does.
What we actually built (and what we didn't)
Armed with these three metaphors, we designed the feature in layers. Each one delivers value from day one. No "wait six months for the full vision" roadmaps.
Layer one is session mobility. Pick up your AI conversation from your phone. This already exists — Anthropic shipped it last week. You start a session on your laptop, grab your phone, and continue right where you left off. No building required.
Layer two is the proactive notification layer. Your AI reaches you with what matters. Think of it as a pager. Not a full conversation — just: "Your 2pm prep is ready" or "That research you requested found something interesting." This is the executive assistant behaviour. It comes to you.
Layer three is the full messaging interface. Dispatch tasks, get responses, have real conversations — all from a chat window. This is the destination, and layers one and two mean you're getting value long before it arrives.
The thinking matters more than the tech
I could have written this entire article about WebSocket protocols and API architectures and message queuing systems. That would have been impressive, perhaps. And useless to you.
The three metaphors are what I actually want you to take away. Not because they're specific to AI messaging. Because they're a thinking process you can apply to any feature, any tool, any shiny new capability you're considering for your business.
Executive assistant, not help desk: Are you building something reactive or proactive? Is it waiting for people to show up, or is it meeting them where they are?
Walkie-talkie, not control room: What context is this tool for? Strategy or execution? Deep work or quick coordination? Don't build a control room where people need a walkie-talkie.
Concierge with multiple entrances: Can different people access this capability through doors that feel natural to them? Or are you forcing everyone through the same lobby?
The sexy feature was "text your AI from your phone." The real feature was "an assistant that understands how founder-brains actually work, delivered through a door you'll actually walk through."
The thinking tools got us there. The technology just followed instructions.
Ed
Source
Ed Dale voice memo/dictation on planning instant messaging for AI assistant system, February 2026. Based on research from [[Outlier Swarm - Messaging Interface for Claude Code - 2026-02-25]] and [[Research Swarm - IM Interface for Claude Code - 2026-02-25]].