Jev caught every meaningful chart in our benchmark. $0.06 per 1,000. 0.14 seconds. Last on precision. Fast and cheap enough for every branch point. The accuracy still has to be watched. With @ramsoma:

We put GPT-6 Sol, GPT-6 Luna, and Claude Opus 5.5 through our Amplitude Global Chat eval and mapped the outcomes on our quality/cost frontier. Our eval is 185 analytics tasks from simple to complex: building a chart, editing one, choosing the right event in a taxonomy, getting the correct date range, finding the property that most explains the variance in a chart. These evals are all based on real customer queries. We grade answers on correctness only. We measure cost and latency separately. Here’s where each new model landed (one run each so±2%): -Claude Opus 5.5: 83% correct at $0.70 per question. That's about 2 points behind our top score (GPT-6 Astra) at roughly a sixth of the cost. -GPT-6 Sol: 81% at $0.58. It scored a little lower than GPT-5.6 Sol, but was considerably cheaper (less than half the price). This is the cheapest model we've tested that clears 80%. -GPT-6 Luna: 76% at under $0.04, making it the cheapest model on the frontier. It's about 3x cheaper than GPT-5.6 Luna and achieves about the same score. The tradeoff is speed: its p95 latency was ~400 seconds, about 3x slower than Opus 5.5. Even with the inclusion of these new releases, GPT-6 Astra still gets the top score (84%) on our eval. The tradeoff is it costs $4 per question and was one of the slowest models we tested (900 seconds at p95).

Building better products shouldn’t come with coordination costs. Wave keeps work moving from evidence to design to code, giving PMs, designers, and engineers more time for the areas that need their judgment.

The code nobody touches. The setup instructions buried in DMs. The tool only its creator understands. Meet the seven deadly sins of DevX. Which one are you?

I’m an engineer building Wave, @Amplitude_HQ’s proactive product agent for helping teams build self-improving products. This diagram gets at what we’re trying to build. The pod sets the outcome, shares the context, and defines the guardrails. Wave senses what’s happening, helps prioritize opportunities, and carries work across product, design, and engineering. Even a small customer problem can need all three, and lining them up is often harder than it should be. Wave can keep the same thread intact from customer evidence to a proposed design to a code change. The PM decides what’s worth doing, the designer shapes the experience, and the engineer owns what ships. What I want is for Wave to take on more of the coordination and execution, so PMs can spend more time talking to customers, designers on user research and design systems, and engineers like me on product architecture and long-term technical decisions. Wave takes on more of the execution load, so product teams can spend more time where their judgment matters. That’s the kind of agent I’d want on a team, and it’s what we're building with Wave.

What does Setup & Activation mean when your users are Agents? What are some frameworks & tactics you can use that support both human + agent users? ..we've been exploring these q's at @Amplitude_HQ. We recently moved to a single setup prompt as the unified starting point for both humans & agents. It combines deterministic CLI commands for the steps that must be exact with skills for the steps that depend on your repo. Get started on Amplitude has never been easier 👊

Last week @cxodev and @Amplitude_HQ hosted nearly 300 PMs + execs at literally the hottest happy hour I've ever attended 🥵 The best part? I got to record mini-mini How I AIs with my favorite AI people including @graceclarke @JJEnglert @yourgirlhils 📸 @cosmos_raj

i'm really excited to see more headless products like this release from amplitude one of the coolest parts of everyone going headless: it doesn't just make products easier to use in isolation, it also makes them much easier to use with other headless products prompt: "identify top amplitude events w high correlation to user activation. pipe those into our knock onboarding journey and propose an experiment we can run to lift activation. bring results back into amplitude" in past this meant hours/days of data integration shlep work. the worst part of the job now, as everyone goes headless, it means agents can take on more of that data integration work for us this will unlock so many outcomes that were previously hidden in the fog of shlep

(1/6) I've been a growth PM for over a decade, previously at Dropbox, now @Amplitude_HQ. The Amplitude growth team now runs almost entirely headless.. agents do the work & humans set direction. We're moving FAST. A few examples:

Introducing Headless Amplitude. Say bye bye to the user interface. Agents are now first class users of Amplitude. - All core areas of Amplitude are now accessible via MCP server and we've optimized the interface for minimal token consumption - Developer CLI to manage events, experiments, flags, and projects so you can make changes in Amplitude deterministically - Service accounts so bots and scripts can query Amplitude on their own accounts In August alone, 6.3M MCP tool calls ran through Amplitude, 5x what we saw in March. Querying product data is now our fastest growing agent use case. You shouldn't have to log into the 100th SaaS dashboard to understand what your users are doing. Amplitude can show up in Codex, Claude, Cursor, or whatever agent your team is already using. We’re building toward that future with MCP, developer CLI, service accounts, and agent readable experiences. Use Amplitude without using the UI.

(1/7) my team at @Amplitude_HQ is trusting AI to drive more of our product development and seeing incredible results! Wave is our self-improving product agent: it finds friction, orchestrates the fix, and only ships the stuff that proves impact. read on for more examples of what it's doing in our own product:

HYBRID's (YC21) AI coach botched 1 in 4 chats and logged zero errors. The bug was a tool retry that kept misfiring. The agent still produced an answer every time, so analytics saw a completed session and observability saw a clean trace. The only thing that would have caught it was a conversion drop, far too late and very expensive. Almost every team shipping agents has some version of this blind spot right now. Analytics says a session happened and observability shows the trace. Neither of them inherently show whether the reply was any good, whether the user trusted it, or whether it moved them toward converting. That is the whole business case for the agent, and plenty of teams answer it with vibes and a weekly eyeball of transcripts. Amplitude hit the same gap with its own agents in 2025 and built Agent Analytics for itself. Now it's opening the tool up to other teams. As per Amplitude, every trace gets scored for task success, failure type, and tool-call errors, and every conversation becomes a measurable conversion event. The pitch is you can rank model providers on conversion, cluster conversations by intent, and fix the failing topic that costs the most first. Three customer results came out with the launch. Amplitude says The Economist runs Lens, its AI assistant, on Agent Analytics at a 96.9% task success rate with weekly failures down 84%. HYBRID fixed its 1 in 4, and the same numbers show athletes who use the AI coach retaining 4x more than athletes who skip it. Cashbook (YC21) discovered that a bad first AI answer quietly cost it retention across their app by 15%. Most agents in production have a 1 in 4 hiding somewhere. Reply quality is turning into a metric teams track on purpose, and the ones counting it now will catch theirs before the revenue chart does. No one builds AI agents just for the sake of it, especially as AI is increasingly expensive. Ultimately you're trying to drive adoption and business outcomes. I’d love to hear your thoughts. How are you measuring your agents today?

Dogs deserve self-improving products too. Wave found user friction on Dan's site, prescribed the fix, and sent it straight to Cursor to ship. Now Dan can get back to more important things (naps, zoomies, treats...).

It was awesome speaking at @Amplitude_HQ Design x AI last night to 200+ designers on how we are using AI to build Rovo at @Atlassian - refreshing to see a event around design process, craft and community - thanks for having me!

Quit logging into Amplitude with ChatGPT Data agent. I'm excited for @Amplitude_HQ to be a featured ChatGPT Work plugin and a top source for its new Data agent launch today. This is a whole new way to use Amplitude data to understand what your users are doing and decide what to build next. Now anyone can do that without leaving their chat. Too many SaaS companies are trying to get you to visit their walled garden. We are all in on the headless future here at Amplitude. There's no more need to login to the 100th SaaS dashboard!

Design teams are rewriting how they work, and nobody agrees on what the new version looks like yet. On September 10 we’re hosting Design x AI in San Francisco. Nine designers from @SpaceXAI, @SlackHQ, @evelegalai, @tryramp, @Atlassian, and @Amplitude_HQ will show what's actually working. Almost full, register now: