It was a simple message from one of my coworkers late last May that forever altered our startup’s trajectory: “On June 30 Slack is cutting off the level of data access we need for Hoop to work.”
My stomach dropped.
At Hoop, where I’m a cofounder, we'd spent the previous year building an AI task manager that automatically captured action items from Slack conversations, email threads, and meeting transcripts and put them in one spot so business executives never missed tasks. The product was finding a market: Users loved it, retention was good, and we were adding hundreds of paying customers each month.
The problem was Salesforce. The company, which owns Slack, was clearly building a moat around its data. Without this datastream, we lost the most differentiated piece of our product and the heaviest source of noise where Hoop could help customers find clarity. Our product-market fit survey told us that most of our customers found the Slack connection indispensable. Without it, we were likely to lose them.
Our first reaction was panic. We considered trying to find a workaround like adding Microsoft Teams support or looking for some sort of loophole. But then we realized we had to rebuild and rethink both our product and processes. So much had changed since we began building this product, including the ways we believed people would interact with software as well as their expectations of what software could do for them. We came to realize what every company is going to realize as AI advances: As you integrate AI-first thinking, you stop doing the work and start directing it, and you can operate at a completely different level.
We were going to rebuild the product and our team’s processes in a whole new way.
Becoming AI-Native
Most of us at Hoop came from Trello, which we built from the ground up to a $425 million exit to Atlassian. To successfully transition to being AI-native, we had to question which of our “best practices” and processes still made sense. Instead of injecting AI into existing workflows, we needed to rethink the workflows themselves.
To find our pivot, we followed a four-step process to be systematic about any new direction. Before the crisis, our development cycle looked something like this:
- The product team writes a product requirement document (PRD).
- The design team creates mockups of the product.
- Engineers break down work, create tickets, and get to work coding.
- The marketing team prepares assets.
We’d build, test, deploy. Rinse, repeat. We certainly sprinkled some AI in there, but most of our code was still written by humans.
Step 1: Starting over
The transformation started with every person on our team changing their starting point:
My cofounder Justin began to use ChatPRD—a chatbot for product managers—as a thought partner. He was able to get more in depth on product thinking more quickly, and with more productive revisions with ChatPRD as a teammate questioning his assumptions.
Our engineering team experimented with agentic coding tools like Cursor and GitHub Copilot and eventually Claude Code. It—or rather “they,” as we often ran up to 10 agents working in parallel at a time—became our newest engineers. We'd feed Claude Code our AI-assisted PRDs and get back working prototypes we could iterate on.
It wasn’t easy. Claude Code was fast in some ways, but slowed the team down in others. It took trial and error to figure out the level of fidelity to expect from Claude, review the copious amount of code it generated, and develop a process for how to parallelize work amongst humans and agents.
But this setup made it clear to everyone how critical it was to adapt and upskill.
Step 2: Finding a new direction
We experimented in the open. The team shared their newest insights via a dedicated Slack channel, and we demoed new tools and ideas regularly in meetings. As CEO, I was a deliberate and active participant; I knew it was important for my own work to be visible to the entire company.
Before long, we landed on the next product idea for the business. Time was tight, so we had to be ruthlessly practical. We got together and came up with some guidelines for the next pivot:
- Something we can build quickly
- Either a lot of customers or a relatively high price point
- Plenty of margin between AI processing costs and price
- The product itself could be a growth lever
- Business tool
- Clear workflow that provided immediate value
- AI in a way that is aligned with what AI is good at (i.e., transforming unstructured data into useful artifacts)
- Play to the team’s strengths—design, storytelling, product taste
- Bringing AI capabilities to less technical audiences
The winning idea that emerged was to build an AI-native support tool for startups and business owners that was easy to set up and worked right out of the box.
Step 3: Designing the right product
Here’s how it works: Most support tools (Intercom, Zendesk, etc.) in the market bolt AI agents on top of legacy software. You can’t get value out of the agent until it has enough business context, which requires creating and maintaining in-depth knowledge bases. Most startups and business owners don’t have time for that, not to mention keeping information up to date and learning complicated new software.
Hoop works differently. First of all, there’s no new tool to learn. It works directly in Gmail, drafting responses to common questions without macros, templates, or training.
It does this by scanning the website and support inbox of the company during onboarding to automatically generate an editable FAQ. It also pays close attention to the voice and tone of email responses to make sure future AI-written replies sound exactly as the business would respond. In practice this means that Hoop can be set up within five minutes, and you get back hours in the day by not writing responses to repetitive questions, while your customers still get quick responses.
Step 4: Becoming AI-first builders
There is a nice analogue to Trello with this business opportunity: Trello had translated the process of running kanban boards—previously relegated to developer tools—into a delightful, easy-to-use format that millions of people could instantly understand.
We realized we could do a similar play by taking something that was out of reach before—excellent customer service automation—and use AI to make it easy for businesses of any size to achieve.
The AI-native mindset and building approach made this pivot possible at a speed that would have been impossible even six months earlier. We had an initial prototype, website, and ad campaign to test messaging live within the first week. The first design partner (an early customer who helped shape the product) was onboarded the second week.
The uncomfortable truth
If you’re building a startup, we’re in an exciting moment, but also a precarious one. Capabilities are changing at breakneck speed, and new competitors are popping up overnight. It feels like a full-time job just keeping up with all of this change.
Most startups will face existential challenges like the one we faced multiple times. The companies that survive will be the ones that can rebuild fastest when the ground shifts and the ones whose teams have a never-give-up growth mindset.
Being AI-native isn't about using the latest tools or adding chat interfaces to everything. It's about fundamentally restructuring how you think—and what you think you know—about building products.
With that comes the greatest challenge: retaining your senses of humility and curiosity. Senior teams may have the hardest time with this transformation because we've spent decades building expertise, developing intuition, and perfecting our craft. Accepting that an AI can do parts of our job better than we can can feel like professional death.
But when we embrace AI and start directing it, we are able to spend more time on the things that only we as humans can do. Our engineers can build and test much more quickly. Our designers can illustrate their vision by building a prototype in the middle of a conversation. Our marketers can test every message before it sees a live audience.
Three months ago, our company almost died. Today, we're building faster than ever, in a market we wouldn't have discovered without the crisis. We’re bringing a product to market that couldn’t have existed a year ago both due to model capabilities as well as our team’s mindset and shipping cadence.
The question isn't whether you'll face your own platform crisis. It's whether you'll be ready to transform when you do.
