platform for starting and running a digital business

A platform for orchestrating AI-agents to do work

ROLE

Co-Founder/CEO

YEAR

2023

Project description

Project description

Project description

Comet.Computer was a startup I founded along with Simon Dunkelman after Entrepreneur First. Our vision for Comet was a single workspace to build a business, using LLM’s and agents to get tasks done faster. We thought of it a bit like an IDE, but for building a company.

Over a 9 month period we spoke to hundreds of customers, brought on 20 initial co-design partners and launched some early MVPs. Ultimately, we decided to wind down due to a perfect storm of tough competition, technology hurdles and changing personal circumstances.

Process

Process

Visual design and new metaphors

As it was just the two of us working on it and we had to cover all product and commercial, our cadence was more chaotic than a normal project. Luckily, Simon, the CTO, is a great FE engineer and can quickly translate design concepts to code.

One thing we struggled with when talking to customers was trying to explain how the what Comet was, and how it differed from other tools. We eventually landed on the metaphor of Hats, as in “I’m wearing my marketing hat today”, and users to quickly grasped that they could do different tasks in the same space. We found this a good evolution of the desktop metaphor, in that it enabled users to elevate their level of abstraction from execution (e.g. excel modelling) to orchestration (e.g. finance), handing over the task to the agent while maintaining a supervision role.

Process

Our cadence was more chaotic than a normal project, as there was just two of us working on it and we had to cover all product, as well as commercial. Luckily, Simon is a great FE engineer and can quickly translate design concepts to code.

Visual design and new metaphors

Visual design and new metaphors

One thing we struggled with when talking to customers was trying to explain what Comet was, and how it differed from other tools. We eventually landed on the metaphor of Hats, as in “I’m wearing my marketing hat today”, and users to quickly grasped that they could do different tasks in the same space. We found this a good evolution of the desktop metaphor, in that it enabled users to elevate their level of abstraction from execution (e.g. excel modelling) to orchestration (e.g. finance), handing over the task to the agent while maintaining a supervision role.

Solution

Solution

Solution

Our target user for Comet was someone building a digital business, but that didn't want to hire to scale. To enable that, Comet used GPT4, and later Mixtral 8x22bn, to run chained LLM prompts to execute tasks.

The first task we started with was outbound sales, enabling users to run personalised outbound email campaigns at scale.


Here's how it worked:

Our target user for Comet was someone building a digital business, but that didn't want to hire to scale. To enable that, Comet used GPT4, and later Mixtral 8x22bn, to run chained LLM prompts to execute tasks.

The first task we started with was outbound sales, enabling users to run personalised outbound email campaigns at scale.


How it worked:

Understand customer type

Comet would ask the user the type of customer they were trying to reach, asking clarifying questions until it felt it had enough context to start a search

LinkedIn integration

We worked with a partner to integrate into LinkedIn, giving Comet access to potential customers. Comet would run a search on the customer the user had defined and return a lead list in a xls format.

Define outbound strategy

As it worked on building the lead list, Comet would work with the user to help the define their outbound strategy. This was then used as input to craft personalised, targeted emails for each lead.

Test run and execution

When the user was satisfied with the strategy, Comet would print a test run of 5 emails to 5 leads, so the user could verify and correct any issues. Once they approved, Comet would send emails to all leads, using their LinkedIn profile data, and any other info the user had given to personalise messaging.

Demo video of Comet MVP

Results

Results

Results

We built the MVP of Comet and ran a beta group with a small number of customers as design partners, which gave us a lot of insight into the challenge of building a commercially ready version. At that stage, we took a step back and reassessed, ultimately deciding it wasn't the right moment for us to build this startup.

Navigation

Work

FYLD

Comet.Computer

Fintech

Smth

Navigation

Work

FYLD

Comet.Computer

Fintech

Smth

Navigation

Work

FYLD

Comet.Computer

Fintech

Smth