# Self-Improving Products | Amplitude

A self-improving product gets better on its own. Amplitude reads your data, finds what’s worth building, opens the PRs, and measures the result. Run your self-improving product loop with Amplitude.

Source: https://amplitude.com/en-us/self-improving-products

---

# Build self-improving products<!-- -->

Imagine opening your laptop to a list of ready-to-merge PRs, written by your product.

Amplitude reads your product data, finds what needs fixing (or building), and opens the PRs. You review, merge, and move on. That’s a self-improving product.

[Start for free](https://app.amplitude.com/signup)

## What’s a self-improving product...?

A self-improving product is one that gets better on its own. Every product runs the same loop: build, ship, measure. But measure is where it stalls. Self-improving products close the loop themselves.

Build

Ship

Measure

LOOP TIME

6 weeks

mostly waiting on humans

1 day

and getting shorter

Self-Improving

Give a<!-- --> [custom agent](https://amplitude.com/blog/introducing-custom-agents?siteLocation=self_improving_products_ways_custom_agents_body) <!-- -->a job, and it runs on repeat. Your agent pulls context from Linear, GitHub, and Notion, then opens a PR to fix the issue. Fork one of 20+ agents we’ve already built.

Best for the small jobs that keep coming back: alerting, monitoring, bug fixes.

[Browse the library](/agent-library?siteLocation=self_improving_products_ways_custom-agents)

Best for handing off growth and conversion across a whole product area.

[Learn about Wave](/wave?siteLocation=self_improving_products_ways_wave)

Best for bespoke builds where Amplitude is one piece of your stack.

●

I can see in<!-- --> query\_amplitude\_data that 1 in 3 users drop before setup. Let me examine the relevant session replays to understand what’s going on.

Get(amp\_session\_replay\_info)

└ Reviewed 2,556 session replays (ctrl+r to expand)

I can see in<!-- --> amp\_session\_replay\_info <!-- -->that 847 users got stuck on the empty first screen. Let’s create a guided first-run flow to address this issue.

Add(guided first-run flow)

└ Writing<!-- --> src/onboarding/FirstRun.tsx (ctrl+r to expand)

1

export function FirstRun() {

2

– return \<EmptyState title=“No projects yet” />

3

\+ const { steps, complete } = useOnboarding()

4

\+ return (

5

\+ \<GuidedFirstRun steps={steps} onComplete={complete} />

6

\+ )

✽

Writing... (27s • 425 tokens • esc to interrupt)

\>▮

▶▶ auto-accept edits on (shift+tab to cycle)

Your product talks constantly. Every drop-off, rage click, error log, and agent interaction says something about what works and what doesn’t. Amplitude hears the whole conversation.

## Run the loop from your favorite surfaces

*You get a self-improving product! You get a self-improving product! Everyone gets a self-improving product!* With Amplitude, you can run the self-improving product loop from anywhere you work.

Your in-product home base. Ask any question, set up your custom agent, and review Wave’s product opportunities.

What should we build next?

Analyzing onboarding, replays, and feedback…

1 in 3 users drop before setup.

847 replays show them stuck on the empty first screen.

Opened PR #68472 to add a guided first-run flow, ready for your review.

Projected +18% setup completion.

Up next...

CSV export, requested by 41% of active accounts.

Review Flow

Open CSV export

Ask a follow up...

Auto

[Sign up for free](https://app.amplitude.com/signup)

MH

Michael

2:43 PM

@amplitude

Amplitude

2:44 PM

Review flow

Up next:

CSV export

requested by 41% of active accounts →

Message

Includes:

[Chat it up](/blog/global-agent-in-slack-and-teams)

Opened PR #68472 to add a guided first-run flow, ready for your review:

Guided flow

Flow Report Export.csv

4 KB · requested by 41% of active accounts.

Reply to Claude

Attach

Tools

Voice

[Get connected](/mcp-server)

\</>

Using Amplitude, find our biggest activation drop-off and fix it.

▶ amplitude query\_amplitude\_data 1 in 3 users drop before setup

▶ amplitude get\_amp\_session\_replay\_info 847 stuck on the empty first screen

Adding a guided first-run flow. 2 files changed.

▼ src/onboarding/FirstRun.tsx

+142

-18

...131 more lines

▶ src/onboarding/steps.ts

+34

-6

Commit & open PR

→ PR #68472

## See more surfaces

-

[View MCP setup docs](https://amplitude.com/docs/amplitude-ai/amplitude-mcp)

## You choose your role in the loop

This isn’t the Terminator. You have full control of your self-improving product loop. Set the directions, metrics, and guardrails, then decide how much you want to babysit and how much you let it run. Low-risk changes ship on their own. Anything sensitive waits for your review.

**Everything is safe and secure.**

We don’t train our agents on your data. [Read the FAQ](/ai-agents).

Full control

Human in the loop

You review every change before it continues, and nothing ships without your approval. Most teams start here.

Periodic control

Human on the loop

You set the rules, supervise, and step in when you want to. Otherwise, the loop keeps running.

Minimal control

Human out of the loop

You stay out entirely and let it run. Today, that’s mostly low-risk changes and build-your-own setups.

## Wave is the future of self-improving products

Here’s the dream. Point Wave at your product and walk away. It finds what’s worth building, from quick fixes to new features, and hands you ready-to-merge PRs. Then it does it again. Before you know it, your tiny team is outshipping entire product orgs.

***Wait, so is this real today?*** Parts of it. We’re building the rest with a small group of teams. Come build it with us.

[Become a design partner](/wave)

## We run the loops ourselves, too

We ❤️ self-improving products. Our team has shipped everything from paper cut fixes to bigger product bets.

Agent homepage

+160% card-driven sends

Before

5 days

After

15 minutes

+3% onboarding completion

\~6% completion

\~9% completion

+45 signups per week

+6% searches that surface results

Searches that surface results: \~92%

Searches that surface results: \~97-98%

## Frequently asked questions

A self-improving product is a product that gets better on its own. Amplitude reads your product data, finds what’s worth fixing or building, opens a PR to make the change, and measures whether it worked. Then it does it again in a loop.

, no credit card required. [Spin up a custom agent](https://amplitude.com/agent-library), [become a Wave design partner](https://amplitude.com/wave), or [start wiring Amplitude](https://amplitude.com/docs/amplitude-ai/amplitude-mcp) into your stack with MCP and CLI

Start building self-improving products

You could have ten great PRs waiting for you by your next standup. Only one way to find out…