# Internal AI platform

Built and led an internal AI platform that gives teams across Adly's portfolio per-brand context, connected to the systems they already use.

- Author: Robert Botto (https://robertbotto.com/)
- URL: https://robertbotto.com/work/internal-ai-platform
- Published: 2026-10-04
- Updated: 2026-10-04
- Context: Adly, company-wide, May 2026 to now. I built it and lead it.

## Key figures

- **323**: Merged pull requests, 271 of them mine
- **7,576**: Automated tests
- **482**: Linear issues, 423 of them mine
- **8**: Company-wide AI Office Hours sessions

## Context

Adly runs a portfolio of internet businesses. Each brand's truth lives in different places: P&Ls, Asana, GA4, Search Console, GitHub, ad accounts and Slack. Generic AI tools know none of it.

## The problem

Teams wanted to work AI-first, but an assistant without company context gives generic answers, and one with too much access is a risk. The company needed AI that knows each brand's numbers, respects who can see what, and works inside the tools people already use.

## What I built

1. Brand systems: P&Ls, GA4, Search Console, ads, GitHub, Asana, Slack
2. Per-brand context and access rules
3. Ask in Slack or the app
4. An answer, a scheduled brief, or a draft pull request

An engineer reviews and ships the features teammates ask for.

### The platform

I started it in May 2026 and built most of it myself: **423 of 482 Linear issues and 271 of 323 merged pull requests**, backed by 7,576 automated tests. Every brand gets its own context and access.

### Scheduled intelligence

Reports post to Slack on a schedule, led by a daily leadership brief for one of the brands.

### Community-built features

Proven end to end in Jul 2026: a teammate asks for a feature, the platform opens a draft pull request, and an engineer approves it.

### The habits around it

I started AI Office Hours (**8 sessions, 30 to 35 invitees each**, including leadership), co-lead the AI-first rollout, and administer the company's AI vendor stack.

### Engineering automation

An automated PR review bot on every pull request, and an agent that turns production errors into deduplicated tickets on its own.

## Results

| Measure | Value |
| --- | --- |
| Linear issues | 482 (423 mine) |
| Merged pull requests | 323 (271 mine) |
| Automated tests | 7,576 |
| AI Office Hours sessions | 8 |
| Invitees per session | 30 to 35 |

### Notes

- May 2026 to Sep 2026.

## About the author

Robert Botto is an AI-native product leader and forward deployed engineer. He leads a product team at Adly, a portfolio of internet businesses, builds his own AI products, and advises early-stage and AI-native startups on product, pricing and growth.