Product Marketing @Amplitude
San Francisco
Joined November 2008
- Tweets559
- Following232
- Followers159
- Likes457
Nikhil Gangaraju retweeted
(1/6) i'm the pm on Wave at @Amplitude, our agent for self-improving products. for the past few months we've had it running on our own product. it finds problems, ships fixes, and measures whether they worked. here are some highlights:
A quick write up on how Amplitude fine-tuned our Agent product to match frontier Sonnet performance at 1/3 the cost and how you can do the same with Agent Analytics.
Rewiring Amplitude from a tool you use into a team of Agents you now lead 🤖
Today we're launching Amplitude Custom Agents. Teammates that automate recurring workflows on your product data, 24/7.
The signal your team needs already exists in Amplitude. Getting it to the right person, in the right format, before the moment passes still requires someone to do the assembly.
Custom Agents change that:
- Describe the job once: give the agent a name, instructions, and output format and it runs that workflow on schedule
- Connects to the tools your team already works in: Slack, Jira, Linear, GitHub, and more
- Runs on behavioral data from Amplitude, so the output reflects what your users actually did
- Pushes it to systems where you work: a filed ticket in Linear, a Slack message, a doc ready before a customer call
If you're lacking inspiration, here are a bunch of use cases you can start from: amplitude.com/agent-library
Have fun!
Today we're announcing Wave by @Amplitude_HQ , a proactive product agent that helps teams build self-improving products. Wave continuously analyzes your Amplitude data for signals, surfaces product improvement opportunities that you or your agents can approve and works to ship context defined specs, then tracks the downstream experiments, outcomes, and learnings.
As software development costs drop, we’re seeing the bottleneck quickly shift from how fast you can build to deciding what’s valuable to build. Are users actually retaining, engaging, or churning? What are customers asking for? Is what you’re shipping working? Are you learning from experiments?
For context @AnthropicAI launched Fable 5, a new Mythos class of models yesterday, which ran a migration across a 50-million-line codebase within a day that would otherwise have taken a whole team over two months. This shift has profound implications for how teams build products.
With Wave, every product team can turn the signals buried in your data into the next move worth making, then measure whether it landed. In this new era as Analysis gets cheap. Judgment and proof on data you trust just became your moat.
Learn more at: amplitude.com/wave
Nikhil Gangaraju retweeted
Today we're introducing Wave. A proactive product agent that helps teams build self-improving products.
Every product team runs the same loop. Build, ship, use, learn. AI has made building and shipping extraordinarily fast. Understanding usage and learning still happen by hand.
Wave runs the whole loop:
- Analyzes your Amplitude data for signals across analytics, feedback, session replays, error logs, agent traces, and experiment results
- Surfaces opportunities in the form of full product specs that you or your agents can approve and ship
- Tracks the experiments and outcomes of what you've built so you can start the loop over again
Your product learns what works and gets smarter every cycle: amplitude.com/wave
You can now track the International Space Station in earth orbit and calculate its trajectory to your desired location
Introducing ISSRightNow.app
The ISS travels at 17,000 miles an hour and completes one full orbit every 90 minutes. It passes over 90% of Earth's population and is the third brightest object in the night sky. You can even see it with the naked eye.
With ISSRightNow.app you can now
> Map live telemetry including speed, latitude, longitude, altitude and orbital periods
> Discover city land masses or geological features in the ISS path
>Search or use browser geolocation to estimate upcoming visible ISS passes.
>Show pass times, duration, and direction of travel for points of interest
Built using Codex from @OpenAI , uses @threejs for rendering, fetches public ISS @Space_Station telemetry and calculates approximate future pass times locally.
Feedback, (bugs) or ideas welcome!
We analyzed 20K+ regular Amplitude power users to understand how eval scores on their Global Agent sessions correlated to week over week retention for the usage and found an signal that predicts 3X agent retention. Check out the full post from @vinaygo!
Hey friends! I'm hiring for a PMM lead for Agents at Amplitude. 🤖
4 months ago we launched our Agents product. I co-wrote the story with Jacob Newman of how we built it: lnkd.in/gmdbvD8B
TLDR: most analytics agents just generate SQL faster. They see columns and tables. They return a number that's precise, fast, and sometimes wrong without the right context about your product.
We took a different approach. Amplitude is a semantic model — every chart, funnel, cohort, and experiment that teams create through normal usage becomes context that agents draw on. One customer uploaded a 63-page data dictionary and enriched descriptions for over 5,000 events overnight. Another team noticed the agent was weak on certain topics, built some dashboards, and watched accuracy improve in real time. We went from a 9% baseline (that's best-in-class text-to-SQL on production workflows) to 76% across descriptive, diagnostic, predictive, and prescriptive analytics tasks.
That was Act 1. Here's whats next.
We're now building the next chapter of our Agent story.
Agents inside your workflows — Agents that work where you work i.e Teams, email, Claude, Cursor, Codex. Think headless, composable and AgentUX
Agents that you can design— Custom workflows that stitch together workloads such as improving funnel conversion, error tracking, and retention problems. They push findings and, in some cases, take action on your behalf
Long-horizon agents — Agents that understand your codebase and your objectives. They identify a list of problems and get to work on them. Think proactive analytics (agents that plan, monitor impact and design outcomes ). AI-native teams can run these agents inside or outside amplitude and continuously improve their product.
This is the PMM role. You'd own the narrative for the most ambitious product bet in Analytics today — one that’s already in motion
If you're a builder who wants to market something you can shape, DM me. you'd be working with some of the best in this segment cc Jacob Newman @frankdotlee @willdjthrill @_bgiori Eric Carlson Tommy Keeley and more!
Job description in the comments
Less than 1% of Amplitude insights came from agents using our MCP a few months ago.
That number has grown 20x. And it's climbing.
We just shipped the Amplitude MCP Skills marketplace.
10 years of how we actually run product reviews, audit replays, and score opportunities, packaged into skills Claude, Codex, or Cursor can execute.
5 skills worth knowing:
Weekly Brief — narrative standup comparing current week vs. 4-week baseline, filtered for statistical confidence
Replay UX Audit — watches session replays, clusters friction by severity, cross-references with user feedback
Analyze Experiments — full ship/no-ship checklist: SRM check, 7 validity flags, primary/secondary/guardrail metrics
Discover Opportunities — mines for dropping funnels and stalled features, scores each as a RICE bet (only surfaces findings with multi-source evidence + score ≥100)
What Would @lennysan Do? — pair with Discover Opportunities, get your RICE bets, then ask what Lenny's archive says. Concrete recs, sources cited.
Our PMs, Designers and Engineers @frankdotlee @_bgiori @willdjthrill run these automations weekly to fast-track stands ups, validate assumptions and test multiple hypotheses
Live now in Claude and Cursor's marketplaces. Also on GitHub for Gemini and Codex.
amplitude.com/blog/5-agent-s…
Introducing Amplitude Wizard CLI — it's a single command that sets up Amplitude directly from your codebase. Instead of jumping between docs, the UI, and your terminal trying to figure out what to track, the Wizard reads your code, proposes a tracking plan based on your actual product flows, installs the SDK, wires up Session Replay and Experiments, and verifies events are flowing.
Today’s Wizard handles setup. Coming soon: every Amplitude operation is a first-class CLI command, built for developers and the AI agents working alongside them 🤖
congrats to Cassie Peretore Kaia Rae Peacock William Chin, Michael Bird, Kelson Warner and more!
Getting set up with Amplitude just got a lot easier.
npx @amplitude/wizard
One command. Auth, framework detection, AI-powered event instrumentation, ingestion verification. 18+ frameworks supported.
Try it and tell us what you think. bit.ly/4t5ZlO4
Nikhil Gangaraju retweeted
One of the coolest product demos we've ever launched: Amplitude AI Assistant
A chatbot for your website that can show users how to use your product!
One step closer towards self improving products
Your chatbot searches for help articles. It has no idea this user has been stuck on onboarding for 2 days.
Amplitude AI Assistant is built on your product data, not just your knowledge base. It knows what users did, where they're stuck, and acts on their behalf to get it done.
bit.ly/3OObhWC
#AmplitudeAIAssistant
Nikhil Gangaraju retweeted
We built AI agents that understand what your data actually means using the language of product analytics.
Our agents reason over the semantic context teams create through normal Amplitude usage (charts, cohorts, experiments, taxonomy) and use natural language to help product, data, and growth get to trustworthy answers faster.
What's under the hood: An enrichment pipeline that creates a knowledge graph, organizational context that captures institutional knowledge and our Nova-powered query engine built for agent scale workloads at low cost.
In this article, Jacob Newman and @Nikhilgr
share our learnings from the last 6 months of building and operating agents in production as well as where we’re going with long running specialized Agents for asynchronous work.
Read the blog: bit.ly/3Q3Ovug
