Modem (modem.dev) — Competitor Teardown

Evaluated 2026-09-03. Verdict: do not copy. Steal two pieces. Watch the loop.

What it is

An AI product-manager teammate for dev teams — not a feedback tool. Tagline: "Product execution for the AI era." It ingests support/chat/calls (Zendesk, Intercom, Slack, Discord, Gong), plus eng/product context (GitHub, Linear, Stripe, PostHog, Sentry), auto-clusters feedback with no manual tagging ("real-time discussion graph"), files tickets, delegates coding tasks to agents (Cursor, Devin, Claude Code), and auto-notifies customers when things ship.

Company signals

Founded 2025, Toronto. ~6 people.
Founders: Ben Vinegar (CEO — founding engineer & VP Eng at Sentry), Mike Clarke (CTO — ex-CPO Treasury Prime).
$4.4M pre-seed, Feb 2026, led by Accel + Inovia. Angels: Cohere co-founder Ivan Zhang, Sentry co-founders. Deliberately dev-tool-insider cap table.
Pricing: Free / $80 / $250 by event volume, unlimited users on every tier. SOC 2 already. Shipping roughly monthly (MCP server + public API, Stripe, Zendesk, Notion, PostHog, Automation Templates, Agent Skills).

Their thesis

PM-as-bottleneck in the agentic coding era: coding agents ship faster than humans can triage feedback, so an AI PM does the triage and hands work to the agents. ICP: dev-tool teams already running coding agents (logos: Superwhisper, Kernel, Sentry, opencode).

Why FeatureOS should NOT copy this

Atoms already answered it: one job, one machine, no humanoid PM, no general agents. Modem is the humanoid PM. Chasing them = fighting a funded Sentry-pedigree team on their thesis while the distribution problem stays unsolved. An AI-PM layer bolted onto FOS doesn't fix churn; it adds surface area while the committed wedge sits unshipped.

Where they're weak

No customer-facing surface. No public portal, roadmap, or changelog — no place customers go. Modem scrapes feedback where it dies (tickets, Slack); FeatureOS owns the front door where feedback is born.
Narrow ICP. Teams already running coding agents — early-adopter dev tools. FeatureOS serves PMs at normal SaaS companies.
AI-essay digests. Their recurring digests are generated summary soup — the ignorable Canny-style "here's what happened" model the Atoms doc already rejected.

The real threat

Their "close the loop" pillar — notify customers on ship, personalized release notes from PRs/tickets, recurring digests — is the Atoms compound loop, and they are funded to reach it faster. Modem validates that the loop is the prize. It does not validate their route.

Steal these two (both cheap)

Publicize the FOS MCP server + public API. It already exists. Modem markets "your coding agent can query customer feedback" as a headline; FOS has it and says nothing. One landing page + one changelog post: "Claude Code and Cursor can read your FeatureOS feedback." Pure distribution, near-zero build.
Ship the Atoms digest exactly as specced. Fixed body, paying-customer signal only, no AI essay. Modem's existence is a reason to ship sooner, not to change the spec.

One Thing to Act On

Ship the Monday digest before Modem's digest becomes the default answer to "who tells the PM what mattered this week." That wedge is now contested.