Max is running 20 client interviews to understand AI adoption intent, but without a structured forced-choice question the output risks being a qualitative document that does not drive a product decision. · Seen in: 2026-06-30-1201-magier-okr-planning-session-7crM.md, 2026-06-26-1114-magier-okr-recap-sailing-boat-LMgS.md

A Claude agent that takes a Granola transcript basename from each client interview, extracts the client's stated AI intent, classifies it into one of three buckets (execution partner, AI implementation partner, both), and appends the result to a shared Notion database with a confidence score and the verbatim quote that drove the classification.

Build a /client-ai-intent skill that accepts a transcript basename, passes the full text to Claude with a structured extraction prompt targeting three signals: whether the client wants to do AI themselves, whether they want magier to use AI on their behalf, and whether they are actively building internal AI capability. Output is a JSON record with bucket, confidence (1-3), and quote, written to a Notion database via the Notion API. After 20 interviews the database becomes the input for a pivot-table-style summary that Max uses to make the product decision. The first pilot is the next client interview Max or Sina runs.

The 20-interview sprint is already in the 6-week OKR and the churn rate OKR is failing; if the interviews produce only a narrative document rather than a structured decision input, the sprint will not resolve the strategic question before the next planning cycle.

Designers eyeball spacing instead of using variables, causing the majority of QA comments on OTP builds to be spacing and font inconsistencies rather than functional issues, which extends build cycles and compresses OTP margins. · Seen in: 2026-07-01-1400-peter-x-willi-q-1-1-Kvm3.md, 2026-07-01-1430-abdullah-x-willi-q-1-1-AsZm.md

An n8n workflow triggered when a Figma file is moved to a dev-ready status in Many Requests that calls the Figma REST API, checks whether spacing values on key frames match defined token variables rather than hardcoded numbers, and posts a pass/fail report to the ticket before the developer picks it up.

Build an n8n flow that listens for a Many Requests ticket status change to dev-ready, extracts the Figma file URL from the ticket, calls the Figma API to read style properties on the top-level frames, and compares spacing and typography values against a token reference JSON maintained by Alexa. Any hardcoded value outside a defined tolerance posts a flagged comment to the ticket tagging the designer before the developer starts. The token reference JSON is a one-time setup artifact that Alexa already has the knowledge to produce. Pilot on the next OTP project after Vuranta.

Peter and Abdullah both named spacing inconsistency as the dominant QA failure mode in the same week, and Peter's Vuranta build proved that component-first discipline cuts rework; the upstream design check is the missing gate that makes that discipline enforceable without relying on individual designer behavior.

Max sends written proposals to prospects like Promio and Paymo and separately records a walkthrough video, but the video is a manual step that requires remembering to do it and is not consistently produced across all proposals. · Seen in: 2026-06-29-0930-henrik-deckert-and-maximilian-fleitmann-TbGF.md, 2026-06-29-1429-tomas-bartos-and-maximilian-fleitmann-GuK1.md

An n8n workflow triggered when a proposal document is created or moved to sent status in the CRM that posts a Slack reminder to Max with a pre-filled Loom recording prompt containing the prospect name, the three scope decisions from the proposal, and a one-sentence framing for the video, reducing the cognitive load of starting the recording to near zero.

Build an n8n flow triggered by a CRM deal stage change to proposal-sent. The flow reads the deal name and the proposal document URL, passes the proposal text to Claude to extract the three key scope decisions, and posts a Slack message to Max with the prospect name, the three decisions as bullet points, and a Loom deep-link that opens a new recording session. The message also sets a 4-hour reminder if no Loom URL has been added to the deal by then. No new infrastructure required beyond the existing CRM webhook and Slack bot.

Both the Promio and Paymo calls ended with Max committing to send a walkthrough video alongside the proposal; if the video is not sent, the prospect has only a written document to share internally, which reduces the close rate on deals where the decision-maker was not on the discovery call.

When a PM takes over an active client project mid-flight, critical context is distributed across multiple Slack channels, Miro boards, and ticket comments rather than being consolidated, causing the incoming PM to miss bugs and client agreements that were never written into the ticket. · Seen in: 2026-06-30-1430-nicolaus-x-willi-q-1-1-4OjO.md, 2026-07-01-1430-abdullah-x-willi-q-1-1-AsZm.md

A Claude agent triggered when a ticket assignee changes that reads the last 30 days of ticket comments, linked Granola transcript basenames, and any Miro or Figma links mentioned in the ticket, summarizes the open decisions, known bugs, and client-stated preferences into a structured handover note, and posts it as the first comment visible to the new assignee.

Build a /handover-snapshot skill that accepts a Many Requests ticket ID. Claude reads all ticket comments via the Many Requests API, extracts open decisions (items marked TBD or awaiting client response), known bugs (items flagged but not resolved), and client preferences (explicit statements about what the client wants or does not want). The output is a three-section markdown note posted as a pinned comment on the ticket. If a Granola transcript basename is linked in the ticket, the skill also reads that transcript and adds a fourth section with any client quotes relevant to the current scope. Pilot on the Blume account where Nicolaus just experienced the gap.

Nicolaus named a specific near-miss where a sticky CDA bug from Miro feedback almost passed QA because it was never written into the ticket, and Abdullah's Telly handover caused Ojas to rebuild elements from scratch; both cases happened in the same week and represent the same missing primitive.

OKR owners track status numbers on Fridays but do not commit to a named next action, causing KRs that require proactive weekly behavior (like client calls and in-person meetings) to stall between check-ins without anyone noticing until the quarter ends. · Seen in: 2026-06-26-1114-magier-okr-recap-sailing-boat-LMgS.md, 2026-06-30-1201-magier-okr-planning-session-7crM.md

A Slack bot that posts to the OKR channel every Thursday evening with one message per active KR owner asking for a single next action and a named day, collects the responses as threaded replies, and posts a compiled digest to the channel Friday morning before the check-in so the meeting starts with commitments already visible rather than being elicited live.

Extend the existing Slack bot infrastructure Willi mentioned in the 1:1 context. The bot reads the active KR list from the OKR database (Notion or the new sprint database Max is setting up), posts a Thursday 18:00 message per owner with their KR name and current status, and asks for one sentence: what is the next concrete action and on which day. Replies are stored and compiled into a Friday 08:00 digest message. If an owner does not reply by Friday 07:00 the bot sends a single DM nudge. The digest replaces the verbal status round in the Friday check-in, freeing the meeting for blockers and decisions only. 1d build on top of the existing Slack bot.

The shift to 6-week sprints was explicitly motivated by the OKR-to-daily-work disconnect named by multiple people in the recap session, and the in-person client conversation KR is at 2 of 10 with six weeks left; a Thursday commitment prompt is the cheapest possible intervention before the sprint ends with the same miss.