Suhit, here is what changed since our call.

You can either watch the video, or scroll through the page. Both tell the same story.

What changed

Last time, we pitched a data layer for GTM tool companies. You asked why we were not leaning into our own GTM expertise. Now we are. Fraktal will help founders build their own GTM tools. One startup has already asked for a demo.

The problem

GTM tools go stale. In ten days, ChatGPT went from citing big news sites in one answer in four to one in forty. A tool that does not track this keeps telling founders to pay for articles on big news sites.

ChatGPT answers that cite big news sites

Slide along the dates. Only the two marked days were measured.

Most GTM tools are not updated with the latest strategies. And most founders do not know what they need from a GTM tool. In our experience working with founders, the loop looks like this. A founder does not know what works on a channel, so they buy a tool. The tool shows numbers and dashboards, but no clear action, and its output looks like every other tool's. Three months later, they switch tools and hope. And the loop starts again.

The loop

What founders need is one tool made for their company. It knows what works now on their channel, and stays up to date. It runs on their own data, shows what to do next, and does it for them.

One tool, made for your company

Knows what works now
Runs on your data
Tells you what to do
Does it for you

What Fraktal is

Fraktal is a data layer that aims to power the next generation of GTM tools. It will work in two ways.

The Fraktal brain. A living record of what works in GTM, built from real results on every channel. It will hold the posts, emails and pages that get replies, clicks and sales. It will refresh every day.

The Fraktal GTM toolkit. A founder will plug it into their own AI agent, such as Claude, Codex or Cursor, through an MCP or API. Fraktal will interview them on the tool they need: for which channel, and with which features. Then their agent, with the help of our pre-made blocks, will build that tool. The blocks will include connectors to their data, data providers, and ready functions such as an AI-writing check.

Example

Build a LinkedIn tool

Add Fraktal and help me build a GTM tool.

Bashnpx fraktal init

✓ Fraktal MCP connected

Fraktal

Which channel do you need a tool for?

SearchAI answersLinkedInCold email

Fraktal

Which features?

Post ideasDrafts in my voiceComment repliesWeekly results

Submit

Writelinkedin-tool/ideas.ts

Writelinkedin-tool/drafts.ts

Writelinkedin-tool/results.ts

Fraktal

Blocks addedConnectorsData providersAI-writing check

Why it will work

Their own data. The tool will run on the data the founder owns, like their calls, their CRM and their customer conversations. So its output will sound like them and use the words their customers use. That removes AI slop, because the tool writes from facts that only this company has.

A signal from Y Combinator. YC's Winter 2027 Request for Startups asks for "LLMs That Write Well". Jared Friedman writes: "As LLMs have gotten better at coding and math, they've actually gotten worse at writing." And: "The same reward signals that make models good at code actively flatten voice and style." Our answer is tools that write from the founder's own data and voice. Source: ycombinator.com/rfs#llms-that-write-well

Always current. The tool will stay connected to the brain. When a strategy stops working, the tool will know the same day, before the founder pays for it. In the ChatGPT example, a tool on Fraktal would flag the change before anyone paid for another news-site article.

Built for one company. A general GTM tool is built for everyone. A Fraktal tool is built for one company's channel, goal and voice. The founder owns it and can change it.

Example

LinkedIn tool

ConnectedYour callsYour CRMCustomer conversations

In your voice

We lost three deals in March because our follow-up came a day late. Here is what we changed.

Strategies Fraktalbrain

✓ Short story posts with one number

✓ Reply to comments within the hour

Engagement podsStopped working today

Where the data comes from

Today, we collect public data from a wide range of sources across the internet. Next, we plan to pay for or license data from winning agencies and existing tools. Later, we plan to add usage data from the tools built on Fraktal. AI will find the patterns. GTM experts will decide what goes into the brain.

1Public datatoday · wide range of internet sources
2Agencies and toolsplanned · paid or licensed
3Usage dataplanned · from tools built on Fraktal

BrainAI: finds patternsExperts: label why, decide

Our moat: we understand GTM at a deep level

Anyone can give an AI agent a prompt. Few people know what a good GTM tool must do on each channel, and why a result worked. That knowledge is our moat. It will go into three places:

The interview. We will write the questions that find the tool a founder needs, even when the founder cannot name it.
The blocks. We will build the pre-made blocks from what works in practice: the connectors, the data, and checks such as the AI-writing check.
The brain. GTM experts will label why each result worked and decide what goes into the brain.

A context layer can store a company's data. A general AI builder can write code. Neither has this judgment, and it is hard to copy.

Proof so far

The ChatGPT change. We read AI search answers about ten of your companies on 16 to 17 October and again on 27 October. ChatGPT answers that cite big news sites fell from 25% to 2.5%.

A startup asked for a demo. On 28 October we spoke with a growth team that runs 30 to 50 AI agents. They tried three AI visibility tools and dropped them: few prompts, few engines, numbers but no action. They want current data that their agents can pull every morning, through an API and MCP. They asked for a demo and for an AI-text check.

Our philosophy, in practice. Matt wanted to buy Supademo, a demo tool with 3,000 customers. Instead, Luka built a custom tool for Ceviche with Claude Code in about an hour. It works better for Ceviche, and Ceviche controls it. That is our philosophy for GTM: a tool made for your company beats a tool made for everyone.

The cost. One AI search reading for one company costs about $1.30 (50 questions, six AI engines).

Is Fraktal another context layer?

No. A context layer knows one company. Fraktal will know what works across the market.

A context layer must keep each customer's data apart, so it cannot pool results. Companies that hold results, like Gong, do not sell them to other tools.

Company	What it does (their words)	Data it holds	To Fraktal	What it does not do
Glean	"Enterprise AI that understands your company." $7.2B valuation.	One company's docs, chats and apps	Neighbor	Show results from other companies
Mem0YC S24	"An intelligent memory layer" for AI agents. $24M raised.	One user's or one app's memory	Neighbor	Track market results
ZepYC W24	Shared context so agents "complete tasks correctly".	One company's data and conversations	Neighbor	Track market results
Octave	"A context layer that learns." $8.4M raised.	One company's messaging	Neighbor	Show results from other companies
Gong	Research from "5,000+ companies". Benchmarks "coming soon".	Sales calls of its own customers	Neighbor	Sell its results data to other tools
Clay	Data enrichment, plus "community-built playbooks". $7.1B valuation.	Contacts, companies, signals	Neighbor	Measure which playbooks get results
Imagine AIYC F25	"Reverse engineering modern B2B growth, starting with LinkedIn."	Public LinkedIn posts and reactions	Neighbor	Measure replies or revenue, cover other channels, or sell to other tools
General AI builders
Lovable	Builds apps from a prompt	The app's code	Possible competitor	Know what works in GTM, or find the tool a founder needs
Replit	AI agent that builds and runs apps	The app's code	Possible competitor	Know what works in GTM, or find the tool a founder needs
Bolt	Builds web apps in the browser from a prompt	The app's code	Possible competitor	Know what works in GTM, or find the tool a founder needs
Fraktal	Helps founders build their own GTM tools, on a living record of what works in GTM.	Public data today; licensed and usage data (planned)	-	Sell one general tool to everyone

GleanNeighbor
What it does (their words)
"Enterprise AI that understands your company." $7.2B valuation.
Data it holds
One company's docs, chats and apps
What it does not do
Show results from other companies
Mem0YC S24Neighbor
What it does (their words)
"An intelligent memory layer" for AI agents. $24M raised.
Data it holds
One user's or one app's memory
What it does not do
Track market results
ZepYC W24Neighbor
What it does (their words)
Shared context so agents "complete tasks correctly".
Data it holds
One company's data and conversations
What it does not do
Track market results
OctaveNeighbor
What it does (their words)
"A context layer that learns." $8.4M raised.
Data it holds
One company's messaging
What it does not do
Show results from other companies
GongNeighbor
What it does (their words)
Research from "5,000+ companies". Benchmarks "coming soon".
Data it holds
Sales calls of its own customers
What it does not do
Sell its results data to other tools
ClayNeighbor
What it does (their words)
Data enrichment, plus "community-built playbooks". $7.1B valuation.
Data it holds
Contacts, companies, signals
What it does not do
Measure which playbooks get results
Imagine AIYC F25Neighbor
What it does (their words)
"Reverse engineering modern B2B growth, starting with LinkedIn."
Data it holds
Public LinkedIn posts and reactions
What it does not do
Measure replies or revenue, cover other channels, or sell to other tools
General AI builders
LovablePossible competitor
What it does (their words)
Builds apps from a prompt
Data it holds
The app's code
What it does not do
Know what works in GTM, or find the tool a founder needs
ReplitPossible competitor
What it does (their words)
AI agent that builds and runs apps
Data it holds
The app's code
What it does not do
Know what works in GTM, or find the tool a founder needs
BoltPossible competitor
What it does (their words)
Builds web apps in the browser from a prompt
Data it holds
The app's code
What it does not do
Know what works in GTM, or find the tool a founder needs
Fraktal
What it does
Helps founders build their own GTM tools, on a living record of what works in GTM.
Data it holds
Public data today; licensed and usage data (planned)
What it does not do
Sell one general tool to everyone

Our ask

Introduce us to more of your founders. We want to talk with them, learn about their needs, and help them build their own personalized GTM tools.

From here

Let's discuss next steps on our next call.

luka@trytimothy.com