You cannot hear yourself at work.

Every other surface of modern life has a mirror. Sleep has Oura. Strain has Whoop. Money has Copilot. The way you sound across a week of meetings — the thing that decides whether your team trusts you, whether the deal closes, whether the room follows — has nothing. You finish a Tuesday and you genuinely do not know if you talked too much, sounded warm, sounded scared. Nobody does.

What gets measured today, and what doesn't

The unspoken version

Everyone built Strava. The right model is Oura.

Every consumer AI product launched in the last three years has been built around the assumption that more engagement equals more value. Push notifications, streaks, gamification, social leaderboards. That is the wrong reference class for this. Oura makes more than half a billion in ARR and has never once nagged you to sleep more. It just shows you the score. Nobody has built that for how you sound.

Two product philosophies

What nobody else has acted on

• Voice AI got cheap enough to run ambiently in 2024. Nobody built a wellness-shaped product on top of it.
• Anthropic's Claude is finally non-sycophantic enough to hold a mirror without being a cheerleader. This was not true in 2023.
• The cultural backlash against attention-extractive tech now rewards anti-engagement positioning. Intentional tech is mainstream.
• The Humane and Rabbit corpses scared investors off hardware-first plays — which means the software-first sequencing we want to run is uncrowded.

Seda.

Seda is a Mac app and an iOS companion that listens across your meetings with consent and silently builds a weekly portrait of how you communicate. Four dimensions to start — balance, clarity, warmth, presence. It opens on Sunday morning. You read it in two minutes. It closes. That's the loop.

The four dimensions

The product principle

Listen, distill, mirror.

Three layers and one rule. The rule is that raw audio never leaves the device. Capture happens on-device with explicit per-meeting consent. Diarization and feature extraction happen locally. Only the derived signals — pace, sentiment, talk-time ratios, distilled phrases — go to the cloud for the weekly synthesis. You can delete any week with one tap and we keep nothing.

The stack

What we do not do

• We do not store raw audio in the cloud, ever.
• We do not surface real-time overlays during meetings.
• We do not send notifications outside the Sunday read.
• We do not score other people on the call. Only you.
• We do not sell, share, or train foundation models on user data.

A $4-7B category by 2028, built bottoms-up.

I am not going to wave at the $200B wellness market or the $50B corporate L&D market. Those are red herrings. The honest math starts with the ~95M US knowledge workers, narrows to the ~30% in roles where how they sound shapes outcomes, and asks what a meaningful slice of them would pay for a quiet mirror.

The bottoms-up build

Comparable anchors

Four shifts that all happened in the last 18 months.

This product was not buildable in 2023. The voice AI cost curve had not crossed the consumer line, the models were too sycophantic to mirror honestly, the cultural mood still rewarded engagement, and the hardware graveyard had not yet scared investors into funding software-first plays. All four flipped in 2024.

The four shifts

Everyone is solving a different job.

There are eight companies people will mention when they hear our pitch. None of them are building this product. Some are practice tools, some are live correctors, some are transcript engines, some are human coaches. The job we are doing — quiet, asynchronous, weekly mirror across the real workday — is uncontested.

Who's near, and why we're different

Consumer subscription, hardware as a moat.

Two-tier consumer subscription at launch. Enterprise seat-based pricing as the second pillar, modeled on Modern Health and Spring Health distribution into companies as a benefit. Hardware in year two as a margin-positive accessory, not a hero product. The math is built on Oura and Whoop, not on a SaaS multiple I made up.

Pricing structure

Unit economics, honest

Sunspell is the right wrapper for this bet.

Seda is the highest-variance bet in the Sunspell portfolio. It only makes sense inside a venture lab that can hold a long horizon and run software-first sequencing without flinching. I have spent the last decade building consumer products for people who do not want more apps. David runs commercial and judges fit better than anyone I have worked with. The lab around us means we are not betting the house on this one card.

The two seats at the table

Why the lab structure matters here

• Seda's right time-horizon is 4-6 years. Standalone venture funding pushes you toward 18-month exits.
• The other three ventures in Sunspell are B2B and cash-generative. They fund the patience this product needs.
• Sunspell already owns the design language, the privacy posture, and the press surface this product depends on.
• If Seda gets acquired, the lab structure makes the deal clean. If it stands alone, the lab gives it air cover for the years it needs.

Pre-launch. Architecture locked. Cohort warming.

We are honest about where we are. The product is in private alpha on Mac with a hand-picked founder cohort. The privacy architecture is decided and built. The brand and design language are live. We are not claiming revenue. We are claiming that the hard, unglamorous parts of starting a consumer AI company — the consent flow, the on-device pipeline, the voice the synthesis writes in — are done.

Where we are today

What the alpha cohort is telling us

• The Sunday morning read is the moment. Users describe it as 'looking in a mirror you didn't know was there.'
• Warmth is the dimension people did not expect to care about and end up caring about most.
• Nobody asks for live overlays. Several have asked us to send fewer signals, not more.
• Three alpha users have asked when they can buy this for their team. That is the enterprise wedge surfacing organically.

Software first. Cohort second. Hardware only when we've earned it.

The sequencing is the strategy. We will not ship hardware until the software has product-market-fit on its own. Every quarter has one job. If a quarter's job slips, the next quarter's job slips. We are not running a press calendar.

Quarter by quarter

The quantified surface for the rest of you.

If Seda works, it does not just become a successful consumer subscription. It becomes the first piece of infrastructure for a category that does not exist yet — the quantified self for the parts of you that are not biometric. How you speak. How you listen. How you write. How you decide. Communication first because it is the surface with the highest leverage on the rest of your life.

The horizon number

The thing we are actually building

What gets unlocked if this works

• A new wellness category that sits between Calm and Grammarly and is larger than either at maturity.
• A platform-adjacent acqui-target that Spotify, Apple, or Anthropic has a real reason to want — not as a backup, as a strategic option.
• A defensible behavioral data moat that the underlying voice models cannot replicate by getting cheaper.
• A consumer brand that proves Sunspell's anti-engagement thesis can scale to millions of paying users — which is the proof point the entire lab is built on.