Foundations · Custom builds · Ventures

We made a $1.5bn digital bank AI‑native.

At GoTyme, we changed the culture and helped the whole bank put more than 70 agentic systems into production, growing the AI team from 3 to 37 along the way. Now we bring that to ambitious companies, from culture to operations to ventures we launch together.

agent · onboarding running
Task
Open an everyday account for this new customer.
Identity verified from ID & selfie
AML & sanctions screening cleared
Everyday account opened · •••• 4471
Issuing virtual card & sending welcome
Approve View trace 1m 48s
01The thesis

The future of work is agentic. Software does the work. People direct it.

In every company, the highest-value thing a person does is shifting from doing the task to running the agents that do it. We help businesses get there.

You

You direct the agents, and each agent runs its own team of agents.

02What we do
Foundations

We make companies AI‑native.

We make people AI-fluent and shift the culture to AI-first. We start at the top, with shareholders and executives, because culture only changes when the expectation of leadership changes.

Custom builds

We build agents into operations.

We build agents for the work that drives revenue, customer service, and operational efficiency, from onboarding new clients, answering customer queries, and chasing overdue invoices.

Joint ventures
Flagship

We co-found new AI companies.

The biggest opportunities today sit where deep domain expertise meets AI. What once took years and millions now takes months and a fraction of that. We pair deep domain knowledge with our coding agents and build a new company together.

03How it works

Agents, built on the company brain.

Every agent runs off a brain that holds your data, your systems and your expertise. It remembers context and decisions, and learns from every action and correction, so it works with the judgement of your most experienced people and gets better the more it is used.

SourcesThe brainAgents
Data Systems Expertise Company brain Sales Support Finance
04Inside GoTyme

We changed how an entire bank works.

When we started, almost no one at the bank used AI. Here's how we changed that.

01

Top-down workshops

We ran AI workshops from the top down, starting with the board and executives.

02

A standard for everyone

Every employee had to meet it, including building their own AI agent.

03

Agents into production

A specialist team shipped agents into the worst back-office bottlenecks.

04

Embedded everywhere

Teams moved into every business unit until AI was how the company ran.

20M
customers across two continents
1,600+
people trained to work with AI
3 → 37
specialist AI team, in a year
70+
agentic systems in production
05Who we are

Why we built Descent.

Andrew Pfaff
Andrew Pfaff
Founder
David Sithole
David Sithole
Founder
Nikos Pasion
Nikos Pasion
Founder

We were lucky. We got to lead GoTyme Bank's AI transformation end to end, taking the bank from almost no AI to deeply AI-native: 37 AI engineers and 70+ agentic systems in production. We had real room to experiment, made plenty of mistakes, and learned an enormous amount.

What stayed with us was the distance. The gap between people who use AI well and everyone else has become enormous. The top 1% of engineers now ship 46x more code than the median. It started in software, but it's spreading fast. Anthropic, worth hundreds of billions, runs its entire growth marketing with one person.

The businesses that close that gap will get two things right: people who can direct teams of AI agents, and data and systems underneath that compound as the models improve. That's the work we did at GoTyme, and it's the work we now do for other companies, faster each time.

And the same shift opens something bigger. Pair deep domain expertise with AI and you can build whole new companies in months. Doing that alongside the people who know a market best is what excites us most. It's why Descent exists.

The name

Why we're called Descent.

We chose the name for two reasons.

1

Gradient descent is how almost every modern AI model learns. With each step, the model tunes itself to make its errors a little smaller. Plotted on a graph, each step moves downhill toward the point where the gradient flattens to zero. How it works ↗

2

The only way to truly become AI-native is to descend knowledge and expectations from the top down: from shareholders to the board and executives, and on to every employee.

Become AI‑native.

We've done it inside a bank. We'd like to do it again.