Jacob Newman

forward deployed engineer just means the guy is not fucking autistic

@ollysmyth_ computa, make this guy funny as hell

@JacobNewma22350 If only I could make my coworkers laugh at my jokes

worst part about my human coworkers is my inability to fine-tune their personalities amplitude.com/blog/agent-per…

(2/2) And the team is just getting started! Spoiler for next week: The practical difference between your data analyst coworker and an Amplitude agent is access to docs, codebase, DWS, backlog, etc... That changes next week 🙌

(1/2) Agents in @Amplitude_HQ just got a major upgrade: long-term memory, voice-to-text, image upload, searchable history, and an audit hub for admins to monitor quality and build trust.

@Amplitude_HQ @SlackHQ @Microsoft Slack and Teams just got 10x smarter 🧠

Meet your 10x data analyst — Amplitude Global Agent now in @SlackHQ + @Microsoft Teams 🚀 Ask your toughest product questions and get linked charts, product analyses, and recommended actions, all in the same thread. See what it can do → bit.ly/4b3Vv0K

@aakashgupta @frankdotlee @Amplitude_HQ this guy @frankdotlee is on to something 🧐

Claude Code + MCP = Vibe PMing Here's your complete guide with @frankdotlee, Principal AI PM at @Amplitude_HQ: 3:45 - Setting Up Claude Code + MCP 11:08 - Top 5 Use Cases for PMs 40:35 - Biggest Mistakes

@willdjthrill @twanlass you mean the worm?

2/ the animated thinking `rider` ™️ with custom :Amp: logo loader from the talented @twanlass

in case you missed it, @Amplitude_HQ dropped the world's greatest analyst agent last week that can leverage the entire Amplitude platform. anything about your data you've ever wanted to know - just ask :)

@Nirmal_Utwani further proof the only skills worth developing in 2026+ are critical thinking & discernment & comfort in ambiguity

@aakashgupta "The PM role got concentrated" +1 PMs have gotten lost in the ever-expanding ocean of "not the hard problem" stuff -- enablement, specs, standups, process, etc. The job was always meant to be "what should you build?"

Spenser Skates just named the constraint that will separate winning product teams from everyone else in 2026. When coding is automated, the feedback loop becomes the bottleneck. Which features drove retention vs vanity engagement. Why the cohort churned in week three. Where the funnel broke in a way no dashboard surfaced. Judgment about what to build requires behavioral data, and behavioral data requires a feedback loop that matches the speed of the build loop. Here's where the math gets uncomfortable. If your agent ships a feature in 4 hours but your analytics cycle runs 2 weeks, you'll be 84 iterations deep before you know iteration one was wrong. The feedback latency that felt merely annoying when teams shipped monthly becomes catastrophic when agents ship daily. This is what Amplitude is actually solving. Their Global Agent hit 76% task success on complex production queries over 6 months - 7x better than text-to-SQL models. MCP connects behavioral context directly into Cursor, Claude, GitHub, and Figma so agents operate on real user data, not assumptions. And they're including all of it with every plan, including free, while competitors charge $20-50/seat premiums for AI features. The PM role got concentrated. Less time writing specs for features that could have been built anyway. More time on the question that was always the hardest one. Agentic analytics is the missing half of agentic building. One without the other is just expensive guessing at scale.

@JustAnotherPM Yes! AND Amplitude won't stop at "Insights come to you" -- in the near future, the same agent that discovered the insight can take the next-best action whether that's delegating a bug fix to coding agents, launching in-product guides, or modifying an experiment

Think about how product analytics works today: Something breaks → someone notices → you go investigate → you find the answer (maybe) → you act on it (eventually) Every step is manual. Every step is reactive. That's about to change.

He's right. And this is the part most product managers miss. Building is easier than ever. The hard part was never building. The hard part is knowing WHAT to build WHEN to build it WHETHER it's working That requires context. And most AI tools have none.

There will be two and only two types of responses to announcements like this: - "I wouldn't trust agents to understand my data // I'm worried about inaccuracies // This use-case needs more time to mature" - "Let's fucking go" 2026 will not be kind to the former.

@willdjthrill when you buy a $500, 6kg stainless steel replica of Frostmourne you gotta find ways to use it (not based on true story)

how it feels to use codex 5.3 to update copy

@_bgiori @cursor_ai @Amplitude_HQ "Skiing is actually a great activity to parallelize with cloud agents" -- 100% brand new sentence

Started using @cursor_ai cloud agents on my phone and was inspired, so I built a mobile friendly version of @Amplitude_HQ's chat agent this weekend on my phone while skiing. A few thoughts: 1. Cursor's cloud agents are great, but missing some features compared to desktop. Specifically the modes (plan, ask, debug), reverting to previous messages & states. 2. No way to transfer the cloud agent to my local machine. After getting home I wanted to clean up some of the slop and do a small refactor using all the IDE features. I ended up having the cloud agent create a summary of our chat and starting a new agent locally. Would be great if this was built-in. Similarly would be great to take a local chat and push it into the cloud to bring it with me on the go. 3. Opus 4.6 is significantly better than 4.5 and is very, very good for vibecoding. Can't wait to try Codex 5.3. 4. Amplitude has a Github action that will build preview versions of the app based on your git branch. This was a game changer, as I could validate my changes as Cursor pushed to the branch without leaving my phone. Every company needs to add this to their frontend CI. Overall super impressed by Cursor's cloud agents, especially with the new models. If you're using cursor you should try it out. Skiing is actually a great activity to parallelize with cloud agents since you can test and prompt while in line and on the lift and let the agent run while you ski the run.