When I joined Teameight, there was no product yet — just a founder’s vision for an AI system that could sit inside sales meetings, understand deals, and help reps in real time.
As the sole product designer and first employee, my role started with turning that vision into something real. But as we shipped an MVP, onboarded design partners, and watched customers use the product, the challenge evolved.
I went from defining the foundation of the product, to figuring out how humans should interact with autonomous AI, to uncovering the problems customers couldn't name and turning them into products they'd actually adopt.
This is that evolution.
The product today — supporting sellers across the full meeting lifecycle:
PRE-MEETING
Prep and account research
IN-MEETING
Live guidance, objection handling, expert AI agents
POST-MEETING
Recaps, follow-up, CRM updates, deal management
01
THE SITUATION
A founder's vision and nothing else. No interface, no design system, no workflow to build from.
WHAT I DID
Turned the founders' vision into an MVP, grounded in domain research and built within the limits of platforms we didn't control.
Starting With the Seller
Before designing anything, I needed to understand the world I was designing for. I talked with people who worked in sales about their day-to-day: where they spent unnecessary time, where human judgment mattered most, and where AI could realistically help. And I studied adjacent products already solving pieces of the problem, particularly AI note-takers and CRMs, to see the landscape we were entering and where we could stand apart.
I combined that research with the founders' vision to shape the first Teameight experience.
Defining the Core Loop
Rather than treating meetings as a single moment, I designed around the full lifecycle of a sales conversation:
Pre-meeting Prep
Live Meeting Assistance
Post-meeting Follow-Up
Prep notes gave reps the context they needed before a call. The in-meeting experience supported reps with live coaching and objection handling while the conversation was happening. Once the meeting ended, Teameight processed the conversation into a recap, action items, and follow-ups. Dashboards and analytics views came later, once real usage showed us what reps and leaders needed beyond a single meeting.
SIDEBar — DESIGNING INSIDE A PRODUCT WE DIDN’T OWN
Building on top of Zoom and Google Meet
The in-meeting experience came with an unusual constraint: Teameight had to sit on top of platforms we didn't control. Patterns that worked perfectly in a standalone product weren't always possible inside someone else's meeting window.
So I worked closely with engineering early on to learn where those limits actually were, and designed within them instead of finding out later and having to backtrack.
Within 4 months we had an MVP and started onboardinig our first design partners!
First full meeting flow in action, from pre-meeting prep to mid-meeting facilitation to post-meeting follow-up.
Getting design partners to commit told us that there was real interest in what we were building. The core idea worked well enough that teams wanted it inside their actual sales calls. So the question stopped being whether the product was viable — and became whether reps would be willing to adopt our product in the highest-stakes moment of their job.
02
THE SITUATION
Reps have almost no attention to spare mid-call, but several autonomous agents could all compete for it at once.
WHAT I DID
Designed the copilot around intent, so it knows when to help and when to hold back, preserving the rep's agency.
When design partners started using Teameight in real conversations, they surfaced a problem prototypes couldn't: a rep has very little spare attention on a live call. They're listening, thinking, responding, reading the room, and steering the conversation all at once — and every piece of AI assistance competes with that.
Redesigned in-meeting experience: timed recommendations, plus opt-in advanced features so new users start simple.
THE COPILOT
Battle Cards
Surfaced when a rep encountered a difficult objection.
Expert Agents
Brought in deep product, technical, or security knowledge.
In-Meeting Copilot
Let reps ask the AI directly for help during the conversation.
As the system became more capable, the design problem became harder. A product agent, security agent, battle card, and rep request could all demand a user's attention at once.
The challenge wasn’t simply deciding what AI should surface.
It was deciding when AI deserved the rep’s attention.
Teameight’s agents run autonomously — they decide when something’s worth surfacing without waiting to be asked. But autonomy without agency erodes trust: interrupt a rep too often and they’ll tune the Copilot out, yet make them manage it manually and you’ve defeated the point of an autonomous system.
Designing Around Intent
We created different behaviors based on who initiated the interaction.
REP-INITIATED
A rep asks a question or triggers the Copilot with a hotkey
→
The response persists until dismissed. The rep explicitly asked for the information, so the system assumes it remains useful.
AGENT-INITIATED
An agent autonomously identifies something worth surfacing
→
The recommendation appears briefly, can be dismissed, and clears on its own. Offered, not forced.
The distinction seems small, but this change did two things. It fixed the clutter. With timing controlling when each agent could surface, recommendations stopped piling up and fighting for the rep's attention. And it set an interaction model for the whole product. The more autonomy the AI has, the more deliberate we have to be about keeping the human in control.
Getting this right had a measurable effect. Between our MVP and the redesigned version that cleared the clutter and stopped overwhelming reps, active use of the in-meeting experience doubled across our design partners.
2×
in-meeting platform usage from MVP to redesign
Sidebar — WHEN THE DESIGN DECISION IS ALSO A BUSINESS DECISION
Designing around the cost of AI
Not every design constraint is about the interface. With AI, one of the biggest is cost. Every time an agent is updated, it has to be reprocessed, and each reprocess carries a price.
Auto-saving every edit to an agent would mean paying to retrain it on every keystroke. So instead of saving silently in the background like most of our system, I added an explicit confirmation step for this workflow. Changes commit only when the user deliberately saves, batching their edits into a single reprocess. A tiny interaction choice, shaped entirely by what the system underneath it costs to run.
So far this had all been about the live meeting. Teameight's web app had a harder problem to crack. We needed reps to actually manage their deals in our platform, but Salesforce had been their home base for years.
03
THE SITUATION
Design partners kept defaulting to their legacy CRMs, like Salesforce and HubSpot, while piloting us, but they couldn't pinpoint exactly why.
WHAT I DID
Used interviews, observational testing, and competitive analysis to find the friction, then redesigned the accounts and opportunity pages to earn their buy-in.
This problem didn't come as a feature request. It came as a behavior. Teams had the green light to pilot Teameight, everyone was onboarded and had access to the web-app. But when it came to managing accounts and deals, everyone kept defaulting to their legacy CRMs.
Talking to Users
To understand the hesitation, I started with interviews. They didn't get me far. Users couldn't pinpoint what they disliked about the legacy tools, or what they wanted from ours. It was up to me to figure it out and show our users what they wanted. So I stopped relying on what they could articulate and started watching what they actually did.
Studying the Competition
Even though those interfaces were dated, dense, and hard to move through, reps still chose them over Teameight. I studied the platforms themselves, not to copy them, but to understand why reps trusted them. These tools are what our audience already knows, and familiarity carries real weight when it comes to trust. Looking closer, the information architecture mostly held up. Deals, accounts, and projects sat clearly at the top level, giving reps a sense of organization at a glance. But that's where the good stopped. The pages themselves were dense, repetitive, and reliant on manual upkeep, forcing reps to hunt for what they needed.
Watching Them Work
Knowing the gap existed was the first step. To see it play out, I sat with our design partners as they worked through their workflows in Salesforce and HubSpot. It confirmed the pattern, dense pages that made reps hunt for what they needed, and gaps where a rep hadn't gone back to log notes from a past meeting.
Two gaps stood out. One was UI, a weak representation of information and hierarchy on the pages themselves, so nothing signaled what mattered. The other was structural, manual data entry baked into how these platforms worked. People pay for good UI, and closing both gaps gave reps a real reason to switch.
THE REFRAME
To earn the switch, Teameight had to preserve what made the legacy tools feel reliable, fix what made them exhausting, and add clear value on top, delivered in a clean intuitive UI.
Sidebar — THE CHALLENGE OF UNPREDICTABLE OUTPUT
Designing for AI-generated content
The hardest part of the redesign was less visible. It came down to the unpredictability of the output, and that wasn't unique to one page type, it ran across meeting recaps, deal pages, analytics, anywhere an LLM was filling in the content. How much the model could find varied every time, some pulled a lot to work with, others had almost nothing. Building fully custom components for each field couldn't keep up with that. The real task became creating a design system that could mold to different outputs, not one built for a single expected shape.
Pricing was a good example. One company might come back with a clean per-tier breakdown, free, pro, enterprise, with per-seat rates. Another might just say "ranges from $5 to $500 a seat, contact sales for enterprise." The same field had to hold both, and everything in between. The interface had to flex around content we couldn't fully predict, not the other way around.
The Solution
The Accounts page is everything about a company you're selling to; the Opportunity page is a specific deal in progress. What makes Teameight's version different is that AI does much of the filling-in. LLMs run account research to populate company fields, and for opportunities, our systems pull from internet research and past meeting data to keep the deal up to date automatically. That meant I had to design components structured enough to make the information scannable and valuable, but flexible enough to gracefully absorb whatever the AI returned. A table, for instance, had to stay readable whether it held three tidy rows or one messy catch-all, structure that made the output usable without assuming what the output would be.
A FEW OF THE CHANGES THAT MOVED THE NEEDLE
(not all changes listed)
The stage progression bar — a clear visual timeline of where a deal stood, front and center on the Opportunity page. Reps could orient themselves in a glance instead of hunting through fields, it made progress feel tangible, and it took a concept they already knew from Salesforce and made it far easier to read.
More color and contrast — I changed the sidebar from white to purple, giving the interface definition against what had been a flat, mostly-white page. A small change that customers reacted to immediately.
Digestible structure — breaking dense information into chunks and using different graphical elements to present things like recent news in a way reps would actually read, rather than scroll past.
Updated accounts and opportunities pages— preserving top level hierarchy, introducing page-level informational importance and progressive disclosure.
The Outcome
Once the changes shipped, reps moved out of their default workflow and into ours. Across the five design partners we had at the time, the CRM workflow reached 100% adoption as part of their stack. To this day, nearly 80% of Teameight users make use of the CRM side of the application consistently for daily deal and account management.
The point was never to replace a company’s system of record. If a team has decades of history in Salesforce, that stays their source of truth, and Teameight pushes and pulls to work alongside it. But for the daily work of tracking deals and keeping accounts current, reps started preferring our UI — it’s easier to use, visually appealing, and cuts down the manual data entry it takes to keep everything up to date.
100%
CRM adoption across our 5 design partners
~80%
of users make use of the CRM consistently
“
The interface and experience are really well thought and designed — it’s so easy to use. The meeting notes are impressively accurate, everything is editable, and the action items make follow-up seamless.
Enterprise Account Executive, Design Partner
“
“You made the navigation experience and platform UI very clear and intuitive.”
Enterprise Account Executive, Pilot Customer
“
After piloting Teameight for a few weeks, our sales team is saving 10–12 hours per week on meeting prep, customer research, and tasks, and our deal velocity has improved by 20%.
Chief Revenue Officer, Customer