What Shipped in Agentic Project Management in H1 2026: A Working Roundup

I was halfway through a Linear sprint review last Tuesday when I realized something had shifted. Three of the tickets in our Done column had been completed by agents — not by anyone on the team. And nobody had remarked on it. We'd just... accepted it as normal.

Six months ago, that would've been wild.

This is a roundup of what actually shipped in agentic project management between January and June 2026. Not what was announced. Not what's "coming soon." Not what someone demoed at a conference and never followed up on. What shipped. What teams can use today.

More shipped than I expected, if I'm being honest. Also more half-baked releases that got quietly walked back. We'll cover both.

January: Atlassian Rovo Agents Go Live

Atlassian had been teasing Rovo agents since late 2025, but January marked the general availability launch. The pitch: assign Jira tickets to an AI agent the same way you'd assign them to a human.

In practice, it's more constrained than that pitch suggests. Rovo agents work best on tickets that link to Confluence docs (the agent reads them for context), have clear acceptance criteria, and don't require cross-repo coordination. But within those limits? It actually works.

A PM friend at a mid-size fintech tried it for their documentation tickets — "update API docs to reflect new endpoint." The agent pulled the existing docs, compared them against the OpenAPI spec, drafted updates, and submitted a PR. Human review took 15 minutes. Manual writing would've taken 2-3 hours.

The catch: Rovo agents are only available on Atlassian's Premium and Enterprise tiers. You're looking at $17.20/user/month minimum before you can even try it. And the agent usage itself has separate API costs that aren't transparent in their pricing page (I dug through three support docs before finding the formula).

Still. Atlassian shipping this first — and not GitHub — surprised me. I had money (not literally, but emotionally) on GitHub beating everyone to market. Wrong.

February: The Quiet Month

February was weirdly quiet. A few point releases:

• Asana added AI-generated task summaries to their Goals feature (minor, but useful for executives who want the TLDR)
• ClickUp shipped an AI writing assistant for docs (not agentic — more like Notion AI)
• Shortcut announced an "AI teammate" beta waitlist (no ship date)

Nothing that changed how teams actually work.

My hot take: everyone was waiting to see how Rovo performed before accelerating their own roadmaps. Classic second-mover anxiety.

March: Google Jules Hits GA

Google Jules moved from limited preview to general availability on March 11th. I've been tracking Jules since their December announcement, and the GA version is meaningfully better than what I saw in preview.

The core value prop: Jules reads GitHub issues, proposes a fix plan, writes the code, and opens a PR. It's not a PM tool in the traditional sense — there's no board, no sprint planning. But it's agentic in the way that matters: you assign work, Jules does it, you review the output.

What changed in GA:

• Multi-file edits — preview was limited to single-file changes, which made it useless for anything real
• Test generation — Jules now writes tests alongside implementation (or you can tell it not to)
• Review requests — the agent explicitly asks for human review before merging, which sounds obvious but wasn't in preview
• Enterprise tier — SSO, audit logs, repo restrictions, starting at $19/user/month

The pricing is interesting. $19/user/month is cheap for what you get — cheaper than Cursor Teams ($40/seat), cheaper than GitHub Copilot Enterprise ($39/user). Whether that's sustainable or an adoption-grabbing loss leader, I don't know. Google has burned money on worse bets.

Jules is solid for isolated tickets. Where it struggles: anything requiring coordination across repos, anything where the "right" answer is ambiguous, anything where the acceptance criteria are fuzzy. In other words — most real engineering work. But for clear, bounded tasks? It ships code.

(I wrote about single-agent limitations back in May — the short version is that tools like Jules hit a ceiling when tasks get complex enough to need multiple specialist agents.)

April: GitHub Copilot Workspace Gets Multi-Repo

GitHub had been promising multi-repo support for Copilot Workspace since their November keynote. It finally shipped on April 8th.

Before this update, Workspace could only operate within a single repository. Which meant anything involving shared libraries, monorepos with multiple packages, or microservices architectures was basically out of scope. Now it can read from multiple repos and propose changes across them.

The execution is clunky. You have to explicitly tell Workspace which repos to include, it doesn't automatically discover dependencies, and the PR workflow creates separate PRs per repo (rather than a coordinated changeset). But it works.

Sort of. I spent twenty minutes wrangling a three-repo update that should've taken five. Still faster than manual, but my expectations were higher.

Real numbers from a team I talked to: they used Workspace to update a shared TypeScript type definition and propagate the change across three consuming services. Workspace found the usages, proposed the changes, opened three PRs. The human review cycle was about 40 minutes total. Manual would've been 2-3 hours plus the cognitive load of keeping context across repos.

For teams already paying for GitHub Enterprise Cloud ($21/user/month), Copilot Workspace is included. That's a meaningful bundle.

May: Linear Ships AI Sprint Planning (Beta)

Linear's been quiet on the AI front while everyone else shipped assistants. Then in May they dropped their sprint planning beta, and honestly — it's the feature I've been most impressed by this year.

Here's what it does: you dump a bunch of issues into your backlog, and Linear's AI proposes a sprint composition. Not random. It looks at:

• Story point estimates (or complexity tags if you don't use points)
• Issue dependencies (which ones need to ship before others)
• Team capacity based on historical velocity
• Upcoming dates from linked milestones

Then it suggests which issues to pull into the sprint, flags risks ("this depends on a ticket that's been stuck for 3 weeks"), and estimates probability of completing everything.

I've been in sprint planning meetings that took 90 minutes and accomplished less than this. Way less.

The AI isn't always right — it underweights human factors like "Alex is burnt out and needs a lighter sprint" and it doesn't know your team's politics. But it's a strong starting point. Better than the blank whiteboard I've stared at too many times.

The beta is free for Linear teams. They haven't announced pricing for the GA version, which probably means they're still figuring it out.

Linear's approach is interesting because it's not trying to do the work. It's trying to help humans plan the work better. Different bet than Atlassian or Google are making.

June: The Acceleration

June felt like everyone decided to ship at once.

Notion added "Notion AI Projects" — agents that can update project timelines, post standup summaries, and draft weekly reports. Still in limited beta, invite-only. I haven't tried it, so I can't comment on quality.

Monday.com shipped AI automations that trigger based on board state. Not exactly agentic (it's rule-based triggers with AI-generated content), but it's a step toward tickets that manage themselves.

Wrike announced a "digital worker" product that... honestly, I couldn't figure out from their launch materials. The marketing copy was heavy on "revolutionize" and light on "here's what it actually does." Pass. (If someone from Wrike wants to explain it to me: I'm listening.)

And the smaller players kept iterating. VeloCalls added agent-to-API logging that's useful when AI tools make external calls — you can see exactly what your agents are doing (useful for call tracking in agentic workflows). JustAnalytics shipped event tracking for AI workflows, so you can measure agent completion rates the same way you measure feature adoption — similar to how you'd track analytics events for any product feature.

What This Means for Teams

Here's my read on the situation as of late July 2026.

If you're already on Atlassian: Rovo agents are worth trying for documentation, testing, and clear-cut implementation tickets. The $17.20/user Premium floor is the real barrier — if you're already there, the incremental cost is manageable.

If you're on GitHub Enterprise: Copilot Workspace's multi-repo support is a genuine productivity multiplier for coordinated changes. Still not good enough for complex feature work, but solid for refactors and dependency updates.

If you want an AI that assigns itself work from your backlog: This is where we're building DevOS. The current crop of tools still require you to trigger the agent. Point it at a ticket, it does the thing. DevOS is pre-launch — we're building toward agents that pick up work from the sprint board autonomously, post standup updates, and request review when done. Built-in Planner, Developer, QA, and DevOps agents working as assignable team members. Planned tiers start at $0 (Free with 2 agents, 50 tasks/month) through $25/user/month (Pro, unlimited agents) and $49/user/month (Team, with SSO and RBAC).

We're not there yet. Still building. But if that direction sounds like what you're waiting for, join the waitlist.

If you're on Linear: The AI sprint planning beta is free and genuinely useful. Turn it on, try it for 2-3 sprints, see if it saves you time.

If you're evaluating this whole space: Start with one tool, one bounded use case. Documentation tickets are a safe entry point — low risk if the agent messes up, high time savings if it works. Don't try to automate your whole sprint on day one. And if you're running paid acquisition alongside your dev workflow, make sure you're protecting against click fraud — automated agents can trigger the same bot-detection signals that fraudsters do.

The Contrarian Take

Everyone's building agents that respond to assignments. You give the agent a ticket, it does the work.

I think that's the wrong frame.

The real unlock isn't "agent does assigned work." It's "agent picks up work, does it, and tells you what it did." The shift from human-triggered to agent-initiated. That's where the productivity gains get weird — where you stop thinking about delegation and start thinking about supervision.

Most teams aren't ready for that. Honestly? I'm not sure we're ready for that either. The tooling isn't ready. The trust models aren't established. The failure modes are still being discovered.

But it's where this goes. And the teams building toward that model now (that's us, that's what DevOS is trying to be) will have a head start when the rest of the market catches up. Or we'll be wrong and feel embarrassed about this paragraph in two years. Both outcomes are educational.

What's Coming in H2

Predictions are cheap, so here are three I'll commit to:

1. GitHub ships Copilot Agents as a branded product — not just Workspace, but always-on agents that monitor repos and propose improvements. They've been hinting at this in their enterprise roadmap.

2. At least one major PM tool (Asana, Monday, ClickUp) ships multi-agent coordination — not just AI assistants, but agents that hand off work to other agents. The single-agent ceiling is too visible to ignore.

3. Enterprise adoption doubles — Gartner's forecasting 40% of enterprise apps embedding AI agents by end of 2026. Even if that's optimistic, the direction is clear.

I'll revisit this post in January and see how wrong I was. Predictions are embarrassing in hindsight. That's why I'm making them public.

Frequently Asked Questions

What major agentic PM features shipped in H1 2026?

The biggest launches included Atlassian Rovo agents (January), Google Jules moving to general availability (March), GitHub Copilot Workspace's multi-repo support (April), and Linear's AI-powered sprint planning beta (May). Smaller updates shipped from Asana, ClickUp, and Shortcut throughout the period.

How do Atlassian Rovo agents work with Jira tickets?

Rovo agents can be assigned to Jira issues like any other team member. They read ticket descriptions, pull context from linked Confluence docs, and can draft implementation plans or code changes. The agent posts updates in the ticket comments and can request human review before marking work complete.

Is Google Jules available for enterprise teams?

Yes, Jules moved from limited preview to general availability in March 2026. Enterprise features include SSO integration, audit logging, and the ability to restrict which repositories agents can access. Pricing starts at $19/user/month for the Teams tier.

What's the difference between AI assistants and agentic PM tools?

AI assistants respond to prompts — you ask, they answer. Agentic PM tools take assignments from your backlog and work autonomously until the task is done, posting updates and requesting review. The agent is the driver; the human supervises. That's the core distinction.

---

Join the DevOS Waitlist

AI agents that work as employees inside your sprints, standups, and tickets — not single-task copilots. Built-in Planner, Developer, QA, and DevOps agents pick up work from the backlog, ship in branches, request review. Multi-model routing (Anthropic, Google, DeepSeek, OpenAI) picks the cheapest capable model per task. Three-tier memory keeps context across weeks, not just sessions. Pre-launch — everything above is where we're headed.

Join the waitlist →