If DevRel's effect cannot be attributed to a click, what four numbers should a program report instead — and is anyone reporting them?
The argument itself is settled and the house position is on record: report the programme in four auditable numbers, and manage the people on craft. Attribution was always the wrong frame — a developer who read the docs, saw a talk, and signed up three months later did not arrive through a link, and building a measurement practice that only counts links measures the wrong thing precisely.
What is not settled is whether anyone can actually produce the numbers.
The instrumentation started shipping
The interesting shift in August was that the tooling stopped being something DevRel teams had to build. GitHub’s Copilot impact dashboard added a return-on- investment section with a named “Passive” seat segment — a vendor voluntarily showing a buyer how many purchased seats are doing nothing. That is the seller side handing the customer an uncomfortable number, which is the same mechanic as the proof-over-adjectives thread, applied to renewals.
Alongside it, two quieter forms of the same idea: docs audits scored against what the support queue actually asks, and crawler-log data putting AI-answer presence on a measurable footing — Meta’s own indexer went from a rounding error to 37.8% of observed crawl share inside a month.
September widened the free tier of that instrumentation. Google’s Search Console AI performance reports went global — impressions from AI Overviews, AI Mode and Discover, though not clicks — and Val Town published the first attributed AI-referral datum any vendor has offered: a self-reported majority of known-source new Pro signups arriving via AI in July. Presence is now cheap to measure; the conversion side of the ledger still has exactly one self-reported entry.
Why it stays open
Because the survey numbers and the tooling numbers disagree. The most recent wave has 62% of programmes reporting to C-level and 18% able to link revenue, with 61% saying they cannot demonstrate impact at all. Better instruments have not yet moved that, and it is entirely possible they will not: the constraint may be organisational rather than technical. The buyer’s side of the table shows the same shape: SlashData’s first engineering-leader wave (September 2026; 2,341 leadership respondents) has 88% of organisations saying they measure AI impact or ROI, and only 39% of those doing it through any formal or automated process. Everyone reports measuring; few have an instrument.
The other thing worth watching is consolidation. The signal-tooling layer has been rolling up — three acquisitions in eight months, all on undisclosed terms. A category whose prices nobody discloses is a category still deciding what it is worth. The largest of the three said what it decided on September 16: Common Room, now Zoom’s, rewrote its home and pricing pages, dropping the “Signals” product menu — job changes, website visits, dark funnel, product-led sales — for “Buyer Intelligence” under a “Revenue OS” framing, with Product Signals surviving as an add-on line in the pricing table. The community-signal category is being renamed toward the sales stack by its own consolidator, which is the practical reason the guide tells you to land raw usage events in a warehouse you control before renting anyone’s signal layer.
Open loops
What would move this thread. Each is meant to be checkable by a date — if one goes quiet, that is itself an answer.
-
A second AI devtool shipping passive-seat reporting, or a public renewal story that cites one. GitHub's dashboard is currently the only vendor putting an uncomfortable seat number in front of the buyer.
by end of Q3 2026
-
A second attributed AI-referral datum, after Val Town's first (2026-09-02) — a self-reported share of new signups, not yet a conversion rate. Presence measurement keeps improving; conversion still has one data point.
by end of Q3 2026
-
A fourth signal-tool acquisition, or the first disclosed price. Every roll-up in the last year closed on undisclosed terms, which is itself a fact about what the category is worth.
On the record
9 dated entries filed onto this thread, newest first. Each carries its own primary source.
September 2026
- 09-16 Common Room — Common Room drops "signals" for "buyer intelligence"
Common Room rewrote its home and pricing pages this week: the "Signals — from anywhere, all in one place" product menu (job changes, website visits, dark funnel, product-led sales) is gone, replaced by a "Buyer Intelligence" menu of Person360, enrichment, Prospector, AI scoring and buying committee. The hero moved from "The AI GTM Platform: complete buyer intelligence, real pipeline impact" to "Revenue OS", which co-founder Linda Lian's post the same day frames as an intelligence layer underneath the revenue stack rather than another app; Product Signals survives as an add-on line in the pricing table.
- 09-14 SlashData — SlashData launches a recurring research line for engineering leaders
SlashData opened a new recurring research programme aimed at CTOs and VPs of Engineering. Its first wave, drawn from the 2,341 respondents in leadership roles in its 31st Developer Nation survey, finds 88% say their organisation measures AI impact or ROI — but only 39% of those measuring use a formal or automated process such as recurring KPI tracking, a dashboard, or automated monitoring.
August 2026
- 08-16 Meta — Meta-WebIndexer's crawl share jumps 17x in under a month
Promptwatch's crawler logs, drawn from over 26 billion data points, show Meta AI's own crawler, Meta-WebIndexer, jumping from about 2.2% of tracked AI-crawler requests to 37.8% in under a month. That 17x surge makes it the heaviest AI crawler Promptwatch tracks, ahead of OpenAI, Anthropic and Google. It fits reports that Meta is building its own web index to stop depending on Google search.
- 08-12 Stripe — Transaction data ties AI vendor growth to localization
Stripe analyzed AI companies' transactions on its platform — 88% of the Forbes AI 50 included — across 35 markets. The 100 largest reach about 120 markets by year three and draw 48% of revenue from abroad. The fastest growers use twice as many local payment methods, which lifted conversion 7.4% and revenue 12%, while adaptive pricing lifted initial conversion 4.7% and subscription lifetime value 5.4%.
- 08-09 GitHub — Adds a return-on-investment section to the Copilot dashboard
GitHub added a "Potential ROI" section to the Copilot impact dashboard. It compares passive, early-phase adoption cohorts against agent-first ones on three numbers: cost per developer per month, that cost as a share of payroll, and pull requests per developer per month. A salary selector recalculates the cost figures live for the chosen pay band.
- 08-06 Sentry — Publishes the numbers from routing its AI bug-fix PRs to Slack
Sentry published a build-in-public post on its internal bug-fix workflow: its Seer agent opens pull requests for detected issues, then Claude routines pick the most relevant engineer and notify them in Slack rather than leaving the PR to be found. It reports roughly 21% more action on those PRs, 13% more 48-hour responses, and 12.5% more close-without-merge. Sentry argues that last number is healthy, since engineers often chose a broader fix than Seer's narrower one.
- 08-06 KnowledgeOwl — A four-gap scheme turns support tickets into a docs backlog
KnowledgeOwl published a method for reading the support queue as a documentation audit. Each ticket gets one of four gap tags at triage: the page exists but nobody found it, the page is unclear or incomplete, the page is wrong on an edge case, or no page exists at all. Support and docs owners then review the tags together monthly or biweekly, and compare ticket volume 30–60 days after a fix ships.
- 08-04 Armature — Product analytics for the sessions your UI never sees
Y Combinator-backed Armature launched on Show HN with product analytics for agent sessions. It wraps your MCP server to reconstruct what agents actually did, the way PostHog or Amplitude reconstruct a human click stream: user intent, agent reasoning, task success and failure. It picked up 40 points, and it is not the first entrant: AgentCat, at mcpcat.io, already sells the same "agent experience" category.
July 2026
- 07-27 The week — The AI seat gets a price, and the empty ones get counted
GitHub priced AI code review as a seat plus a meter on Monday, then shipped the dashboard that counts unused Copilot seats on Wednesday. The vendor selling AI seats is now arming buyers to find the shelfware, because the ROI-skeptical renewal conversation demands it.