🔒DoP Deep goes deeper into the concepts and ideas that are covered in the Weekly Briefing to help you learn lessons from the experiences of top tech companies. If you’d like to upgrade to receive these in-depth pieces of analysis you can upgrade below. New reports are added every month.
DoorDash’s coding agent now automates over 130,000 engineering tasks a month, Starlink’s voice agent deals with 15,000 incoming calls and Amplitude’s CEO said this week that thanks to suggestions from its internal agent, product activation rates jumped from 6% to 9%.
These might sound like vanity metrics in some cases but even if we take them with a pinch of PR-infused salt, the quantitative benefits of new internal agents are definitely starting to emerge.
Asana says its AI team mate led to $100,000 in savings, Ramp’s router cut token spend by 30% and Deel says that its internal payroll agent transformed a process that used to take 45 minutes into just 10 seconds.
It’s been over a year since we last took an in-depth look at AI Agents in practice across multiple categories and based on these examples alone, it’s fair to say that since our last analysis, internal AI agents have moved beyond the experimental phase into scale up mode.
As we return from the summer break again, this new Deep Dive is designed to bring you up to speed with how some of the world’s leading companies are deploying AI agents internally in 2026.
Last year’s analysis looked at 25 different internal AI Agents in practice and for this latest edition it’s doubled to over 50 as more companies use internal AI Agents in all sorts of different, powerful ways.
How this Deep Dive is structured
The Deep Dive includes over 50 different new examples of internal AI Agents in practice from some of the world’s leading tech companies including Stripe, GitHub, Notion, Replit, Atlassian and more.
Here’s a preview of some of the internal AI agent examples included:
The full report is available to download in HTML, csv and markdown format so that you can not only read about agents in practice but actually do something with them; for example, you could upload the markdown into Claude, Gemini or ChatGPT and ask for suggestions on how you might be able to build similar agents in your own product. Sample prompts with ideas on how to do this are included.
Each agent example is broken down into:
Company - the name of the company
Agent / feature - name / explanation of the AI agent used internally
Category - more on that below
What it does - a more detailed description of what the agent actually does, summarized from a product perspective
Reported impact - where possible, a summary of what this agent actually impacted after being shipped internally
The report spans agent examples across both B2B and B2C product teams and isn’t limited to just engineering.
Agent categories include:
Product and Design - agents are now directly involved in the product development process outside of software engineering. The latest crop of agents can dig through user behavior, feedback, and session data to spot problems or opportunities a team might otherwise miss and more.
Engineering / Software Development - agents writing and reviewing code with a human still signing off before it ships. Snapchat, for example, built Casper so engineers can just ask for help in Slack, watch it dig through tickets and code, and open a pull request, all using only the permissions of whoever asked.
Finance, Payroll and Compliance - agents built to handle finance and compliance, reviewing spend, approving what fits the rules and handing the rest over to a real human. Deel’s Akai for example, catches payroll sync failures, tries fixing them on its own, and when it can’t, pulls together the context a human needs.
Customer Support / Sales - agents that are taking live calls and chats. SpaceX has Grok Voice picking up Starlink support calls, walking customers through hardware troubleshooting, and finishing order fulfillment, handling over 15,000 calls a day.
Security - agents are increasingly being used across various aspects of security. Ramp, for example, let AI agents loose on its own codebase with no specific direction, and they found, validated, and patched around 100 vulnerabilities over six days without a single human writing a fix. Other companies are turning to agents to identify security issues too.
Platform & Governance Infrastructure - agents that decide who gets access to what, how agents find the right tool, how everything stays auditable.
A closer look at the agents from each category
Now let’s take a more detailed look at some of the agent examples featured in the Deep Dive, starting with new internal agents built to help teams during the product and design process.