Linear's Agent now lives in 75% of enterprise workspaces and Anthropic just shipped Skills as an open standard — but agents are mute about decisions you only ever made out loud.

The Lead

Your AI agent is only as good as the documents your org bothers to write

Three pieces of news this week point at the same gap. Anthropic shipped Agent Skills as an open standard — directories of SKILL.md files plus scripts that give Claude bounded, repeatable judgment. Linear announced that its built-in Agent now sits inside 75% of enterprise workspaces, with MCP access to whatever else you've connected. Asana pushed its AI Teammates from preview into general workflows. The framing is consistent: the agent does not have ideas. It executes against artifacts.

Which is exactly why most of these rollouts will under-deliver in 2026. The artifacts in most orgs are thin. Decisions are made in a 30-minute meeting, summarized in a Slack thread, and never written down at the level of "options, criteria, owner, by-when." That is the data the agent needs to summarize a program, draft a status, surface a conflict, or escalate a risk. Without it, Linear Agent is a pretty interface to your already-pretty backlog and Rovo is a wrapper around your half-finished Confluence pages.

Today is Friday — the day you usually run the meeting that prepares for next week. Don't. Today's deep dive is about replacing recurring meetings with three written ceremonies that compound: the DACI decision doc, the engineering RFC, and the working-backwards PR/FAQ. They are the substrate your AI tools have been waiting for.

News

Anthropic releases Agent Skills as an open standard

A Skill is a directory containing a SKILL.md plus optional scripts and references — a portable bundle of "how this org does this thing." Now an open spec, so any Skills-compatible agent can use them. Why it matters: the artifact unit of work for AI is shifting from "the prompt" to "the directory." If your TPM playbook isn't a Skill yet, someone else's interpretation of it will be.

Anthropic launches 10 financial agent templates with Opus 4.7

Pre-built agents for pitchbook generation, KYC screening, earnings review, and month-end close. Released May 5 alongside a Microsoft 365 integration and a Moody's data partnership. Why it matters: the "agent template by job function" pattern is the playbook. Expect "TPM templates" — risk register, weekly digest, dependency scanner — within two quarters.

Linear Agent now installed in 75% of enterprise workspaces

Linear Agent connects via MCP to Slack, Notion, GitHub, and others; Skills let teams save and replay AI workflows with one command. Why it matters: if you're a TPM at any scale, the question this quarter is not whether to use an agent — it's which Skills get codified as your team's official ones, and who owns them.

Asana ships AI Teammates and AI Studio, plus smart prioritization

AI Studio automates work behind the scenes via smart rules; AI Teammates are addressable agents living inside tasks; the Rules engine now drives intelligent automation across portfolios. Why it matters: Asana finally has a credible answer to Atlassian Rovo for non-Jira shops — but only for orgs whose project metadata is clean enough to be machine-readable.

Pragmatic Engineer's 2026 AI tooling map: agentic coding overtakes autocomplete

Orosz's recent piece confirms what staff engineers already feel: Cursor, Claude Code, and MCP are the new stack; autocomplete-only tools are losing share. Why it matters: if your platform team's AI strategy still centers on Copilot for autocomplete, you're tracking last year's pattern. Update your platform RFC.

Art of TPM: dependency management is the highest-leverage AI surface for TPMs

"Many companies claim they support AI; in practice TPMs are stuck with generic ChatGPT or Claude while real leverage lives in specialized tooling." Cross-team API deadlines remain the cascade point. Why it matters: if you don't own the dependency-graph artifact for your program, no agent will ever surface the right risks for you.

Tool Spotlight — Linear Agent + Skills

What it is. Linear's built-in agent operates inside the workspace your engineers already live in: roadmaps, issues, projects, cycles. Skills are saved AI workflows — a sequence of agent steps codified once and re-runnable on a single command. MCP access means the agent can pull from Slack, Notion, GitHub, or any custom service you wire up.

How to use it as a senior TPM. Don't start with "summarize the cycle." Start with three Skills tied to written artifacts: (1) RFC-to-Linear — feed an RFC link, generate the parent project plus seed issues with acceptance criteria; (2) cycle-health digest — for each project flag any issue with no recent comment, no assignee, or scope changed without an updated DACI doc; (3) cross-team unblock — given a stalled issue, walk the dependency graph and identify the named human and the missing decision doc. Each Skill should reference an external artifact: the agent's job is to reconcile the artifact with the workspace state.

When NOT to use it. Anywhere the work is "decide whether to do this." Agents are excellent at catching what was decided and tracking whether it shipped; they are mediocre at reasoning under stakeholder ambiguity. Don't outsource the trade-off; outsource the bookkeeping.

Docs: linear.app/docs/agents

Deep Dive — Running Meetings & Async Ceremonies

Three documents that should replace half your recurring meetings

A meeting is a synchronous serialization of a decision that already exists somewhere fuzzy in people's heads. Most senior TPMs run on calendars where 60–70% of the time is in meetings; the typical L7 in big tech sits in 20–30 hours of synchronous coordination a week. The opportunity cost is enormous, and the AI tooling rolling out this quarter rewards the orgs that move that work to written form. Here are the three ceremonies that compound — when to use which, and how to install them without setting off a process-fatigue immune response.

1. The DACI doc — for any decision involving more than two teams. DACI assigns four roles per decision: Driver (frames it, runs cadence), Approver (the single decider), Contributors (named SMEs), Informed (no input rights, only awareness). The deliverable is a one-page doc: what is being decided, why now, options with trade-offs, criteria, the recommendation, the timeline. Contributors comment async; the meeting — if there is one — is 15 minutes to resolve genuine disagreement, not to brief. Atlassian's Team Playbook has a clean version, but the core insight is older: you replace a 60-minute meeting plus three follow-ups with a 90-minute write plus 24 hours of async review. The latency goes down, the audit trail goes up, and your AI agent now has something to reason over. Use DACI when the decision affects multiple teams and the wrong answer costs > one sprint.

2. The engineering RFC — for any architectural change with multi-quarter consequences. RFCs originated in the IETF, were institutionalized at Stripe, Squarespace, Uber, and Airbnb, and are now table stakes at any serious eng org. Format is simple: context, proposed change, alternatives considered, open questions, migration plan, risks. The single most underused field is alternatives considered: if the doc only contains one option, it's not an RFC, it's an announcement. Squarespace's "yes, if" reframe — replace "no" with "yes, if [these conditions]" — is the single best diff to your team's RFC template. As a TPM, you don't write the RFC, but you own three things: the comment-period timeline (typically 5–10 working days), the decision rule (lazy consensus, simple majority, named approver), and the post-RFC link to the work — a Linear project, a Jira epic, or a working-backwards FAQ.

3. The working-backwards PR/FAQ — for any new product, capability, or program-level bet. Amazon's PR/FAQ is a 1-page mock press release plus a longer FAQ section. The press release describes the product as if it shipped: customer, problem, solution, headline benefit, fictional quote. The FAQ section answers the questions a sceptical exec would ask. Bezos's discipline was structural: documents are read in silence at the start of the meeting (15–20 minutes), then the remaining 40 minutes is debate. The point isn't the press release — it's the act of being forced to articulate customer-facing value before any engineering capacity is committed. Use the PR/FAQ when the question is "should we do this at all?" not "how should we do it?" Most senior TPMs underuse it because the eng team would rather start coding; the cost of skipping it is two quarters of building the wrong thing.

Where they fit together. The PR/FAQ scopes the bet. The RFC scopes the technical approach. The DACI doc resolves the cross-team trade-offs that arise during execution. None of them eliminate meetings — they make the meetings shorter, sharper, and tied to a versioned artifact. As an installation tactic: pick one decision a week that would normally have triggered a meeting and convert it. Don't announce a "new process." Just do it once, then twice, then start linking to your docs in Slack. Process gets adopted by example, not edict.

The async-first manifesto reads like a culture statement, but the operational truth is more boring: it's about which artifact is the source of truth for which kind of decision. The TPMs whose programs run on documents are the ones whose AI agents — Linear, Asana, Rovo, Claude — actually have something useful to say next quarter.

Try this week

Pick one recurring meeting on your calendar with five or more attendees. Cancel the next instance. Send a one-page DACI doc instead with a 48-hour async comment window and a named Approver. Track three things: total time-to-decision in hours, count of asynchronous comments, and how many people felt the need to schedule a follow-up call. Repeat with a second meeting next week.

Events

May 31 – Jun 4 · Cisco Live, San Diego — for TPMs running infra programs; the agentic-ops sessions are now the draw.

Jun 1–4 · Snowflake Summit, San Francisco — strong governance and platform-program track if your org runs on Snowflake.

Jun 2–3 · LDX3 London — best single bet for staff+ TPMs in EMEA. Sessions on async ceremonies and platform programs.

Jun 2–3 · Microsoft Build, San Francisco — relevant if your stack touches Azure, Copilot, or Microsoft 365 agent integration.

Late Jun · TechLead Conf Amsterdam — denser than LeadDev, more architecture-leaning. Worth the trip if you're running a platform program.

Reading

Software Engineering RFC and Design Doc Examples and Templates · Pragmatic Engineer — the most useful comparative survey of how Stripe, Uber, Airbnb, and Squarespace structure RFCs.

DACI: A Decision-Making Framework · Atlassian Team Playbook — the canonical version, with template. Fork this, don't reinvent.

The Working Backwards PR/FAQ Process · Bryar & Carr — the operating manual for Amazon's discipline, written by two of the people who institutionalized it.

The things engineers are desperate for PMs to understand · Camille Fournier on Lenny — required listening; the "too many meetings is a PM smell" segment is direct.

The engineering mindset · Will Larson on Lenny (Mar 2026) — pairs well with the Right Hand archetype piece from yesterday.

"PowerPoint-style presentations somehow give permission to gloss over ideas, flatten out any sense of relative importance, and ignore the interconnectedness of ideas."

— Jeff Bezos, 2017 letter to shareholders