R&D · Product Design · UX

Unlocking Human Potential with AI

AI products that make people better at what they do.

Flummoxed is an applied AI research and product studio.

You drew a triangleOpen the portal

Between AI hype and AI doom, there is a shortage of things that just work.
Flummoxed makes those.

HYPE

too much promise

the useful middleTHINGS
THAT WORKwhere we build

DOOM

too much fear

products

Products

Empowering people with products that make them better at using AI...

Tools that work! No prompt engineering, no onboarding, no course.

Deploy

MagicLens

The first product out of the house, and the engine the rest of the line runs on. Its generative workflows and its deploy tooling are what make a second product cheap.

In beta · Momentos launch October 2026

Deploy

Momentos (the invention)

A new category of digital product - a cinematic digital keepsake made from a single photograph. Invented for the MagicLens platform to reach and engage the widest possible audience online.

Launching October 2026 · an invention

Build

Preponder.ai

Preponder is a rare ancient verb that means to outweigh, to be greater in influence or importance. It's our AI-powered media consumption engine that monitors the internet for trends and insights, which we track and use to generate content.

In build · one capability live

Research

Move38.ai

AI news and analysis, powered by Preponder.ai, Move38 keeps the pulse on the AI industry, everything in every category of the market.

Brand settled · nothing built

Research

Gardening app

Demand researched. Nothing built.

concept identity

Research

Vocal coaching app

Demand researched. Nothing built.

concept identity

All products →

why flummoxed

Why Flummoxed.

We chose a very human word for the shared reaction to AI: perplexed, confused, bewildered, sometimes defeated. Being flummoxed asks us to stop, look again and return to our purpose: “Unlocking Human Potential with AI.”

Garry Kasparov faces Deep Blue in 1997 while Lee Sedol studies AlphaGo in 2016, joined by a continuous chess-to-Go board

Read why we called it Flummoxed

the line

Product Pipeline

Every product starts the same way: a problem we want solved, an idea that compels us, and a solution we want to share with our colleagues, friends, and family.

Idea

A problem we want solved for ourselves, or an idea we want to see exist.

Research

Discover mastery through deep subject matter and expert application.

how we check a claim →

Design

Prototype, test, learn, iterate until the experience is intuitive and seamless

04*Studio Pro/Admin · the engine

Build

With an *engine for scale, and PMF for a defined audience we can target.

Deploy

Robust GTM strategy, with community outreach, brand, marketing and PR. Launch is a process, not an event.

We validate PMF, define differentiation through USP/UVP, prioritize opportunity, build, test, learn, optimize and deploy.

an eye on the market

Before we build

01
Can we solve the problem uniquely, is there demand in the market, and can we quantify it?
02
Can we deliver a differentiated product with a unique value proposition, proof points, and a unique selling proposition?
03
Where is there measurable demand for content about our product, and what content can we create to engage that audience?

how we read demand →

founders

2 Founders = 40+ Years Experience

A physicist who builds intelligent systems and a product and growth operator, both shipping products with premium quality and UX.
Between them: 40 years of success and a very short list of excuses.

Jason Schwartz, PhD

CTO · Research & development · Intelligent systems

A physicist who builds the intelligent systems, and the workflow engine every product on the line is made through.

Devin Sawyer

CMO · Product · UX · growth

A product and growth operator who shipped online banking to millions, now building against every gap he found in the generative tools.

Founders →

perspectives

We think a lot about AI!

We are rooted in the history of this technology, and we pay attention to where it is going, including the parts of it that should worry people.

A massive market and an existential shift for humanity!

HistoryOver a Century!

The history of AI traces back to 1913, when Andrey Markov counted letter sequences in a poem and showed that what comes next depends on what came before–a mathematical chain that helped make language computable. That thread runs through Shannon’s information theory, statistical speech and translation, neural word representations, transformers, and instruction-tuned frontier models that now generate, translate and reason across language.

1913Markov chainsA present state changes the odds of the next one.

1948Information theoryShannon measured messages and showed statistics producing increasingly English-like text.

1972Statistical speechIBM used word-sequence probabilities to score what a speaker likely said.

1990Statistical translationBilingual records taught a system translation probabilities instead of hand-written rules.

2003Neural languageLearned word representations began sharing knowledge across similar uses.

2017TransformerAttention modelled relationships across a sequence while training in parallel.

2020–22Prompt + instructionScale, examples in prompts and human feedback made one model useful across many tasks.

Follow the century-long technical thread →LanguageThe human operating system.

Language is how people carry thought, memory and coordination between minds. Today’s frontier AI is built by turning that human system into numbers, then finding faster and more powerful ways to compute what each piece means in context.

1950sRulesEarly systems tried to encode grammar and expertise as instructions written by people.

1980s–90sCorporaLarge collections of real language let probabilities replace many hand-built rules.

2003Word vectorsWords became learned coordinates, carrying relationships a model could reuse.

2017TransformerAttention modelled context across a sequence while parallel training made larger runs practical.

2020PromptsExamples written into a request could guide one model through many tasks without retraining it.

2022Instruction tuningExamples and human preferences made prediction behave more like useful assistance.

See how language became the interface →

Perspectives →

We’re all flummoxed by AI

flummoxed

We’re figuring it out together – one move at a time.