How Amplitude transformed from AI skeptics to all-in

Executive overview

Amplitude spent most of 2022–2024 dismissing AI as hype — engineers resented the grifting narrative, and the models were too jagged to justify a bet. The turning point came in late 2024 when new engineering leader Wade Chambers and acquired startup Command AI showed what was actually possible, triggering a company-wide AI week and a full product pivot.

Traditional SaaS delivery loops — talk to customers, prioritise, build, repeat — break down with AI because customers can't describe what isn't yet possible. Building well requires technology-first intuition, not customer-led roadmaps.

The core insight: AI transformation in an incumbent requires a full year of bottoms-up belief-building before meaningful products emerge.

Why Amplitude was skeptical

• Models in 2022–2023 were jagged: exceptional at some things, terrible at others
• Executives and investors pushing "AI strategy" were asking the wrong question
• Co-founder Jeffrey saw widespread grifting — bold claims, weak capabilities
• Engineers had no organisational mandate to explore; other product lines had clearer ROI

What changed in late 2024

• Cursor and similar tools made AI productivity gains undeniable for engineering
• Wade Chambers joined as engineering leader, bringing bleeding-edge model experience
• Command AI (YC) was acquired; its team had been building model-native products in production
• Together they became the "tip of the spear" for demonstrating what was possible

The AI week playbook

• First goal was not "what should we build?" — it was getting the existing team to use tools and believe in them
• Ran two days of training, then a full hackathon week focused on doing existing work faster with Cursor
• A product leader live-coded a dark mode for Amplitude in front of the whole engineering org — hit a bug, fixed it live; the moment landed
• VP-level participation made the signal clear: leadership was committed, not just cheerleading
• All major AI products that followed — MCP server, AI Visibility, Ask AI — originated as bottom-up hackathon ideas

How the SaaS delivery loop breaks with AI

• Standard loop: talk to customers → prioritise → build → ship → repeat
• Customers can't describe what AI makes possible; they ask for a "faster horse"
• Capabilities are jagged — knowing what falls in which bucket requires hands-on model familiarity
• Product decisions must start from technology understanding, then map back to customer problems
• AI-native engineers often lack domain depth; SaaS veterans often lack model intuition — the best engineers marry both

Organisational changes required

• Two full reorgs of the 200-person product, engineering and design org in one year
• Leaders who were strong in SaaS-mode but not on AI's bleeding edge were moved out
• Acquired three additional YC companies (Craftful, Inari, June) to inject AI-native founders
• Created a dedicated AI team while keeping the core product team running existing roadmap
• Self-selection worked: strong engineers and designers naturally gravitated toward AI projects

What Amplitude is building

• Ask AI: a global chat interface — "cursor for analytics" — letting users query data, pull charts and run analysis conversationally
• AI Visibility: free product that doubled new signups weekly after launch; deliberately given away as lead generation
• MCP server: unplanned, built by one engineer during hackathon week
• Four 2025 priorities: rebuild Amplitude AI-native; make it easier to use; feature parity on non-analytics products; serve marketers to compete with legacy MarTech

On AI Visibility and the startup opportunity

• Visibility features commoditise fast — Amplitude shipped in weeks and made it free
• Real business has to sit downstream: content generation (like AirOps), workflow automation, compliance
• Incumbent advantage: existing revenue base funds giving things away; startups can't easily match that
• Best startup opportunities are vertical-specific agent products tied to a specific buyer, not general agent builders
• Google's B2B execution remains uniquely poor — email, workspace, analytics all vulnerable
• Enterprise AI adoption is blocked by security and compliance concerns — a solvable wedge

Founder mode at scale

• Early-stage founders lead from the front on every hard problem — code, product, customers, people
• At 800 employees that's impossible everywhere; the skill is choosing where to go deep
• Most founder CEOs leave after a decade because the role genuinely requires a different toolset
• Hierarchy has real value: clear ownership prevents the ambiguity that kills execution
• Large company advantage: product-market fit, capital, leverage; the learning shifts to deploying resources well
• "You become the person you always made fun of" — a manager of other people's work — but there's a reason for it

On finding your why and lasting through

• Starting a company is emotionally brutal; at year one or two there's always a rational case to quit
• The successful ones don't quit — that's the number one filter, not idea quality
• Extrinsic motivation (recognition, money) won't carry you through the long stretches of uncertainty
• Get the top node of your goal tree right first; everything else can be derived from it
• "I'll try it and double down if it works" is the worst starting posture — uncertainty requires committed anchoring
• Finding mentors: be crystal clear on what you need to learn, then stay open to where it comes from