DEV Community: Kelly Lewandowski

How my team killed manual standups with Claude + Kollabe MCP

Kelly Lewandowski — Wed, 29 Apr 2026 10:21:21 +0000

The standup didn't die because we hated the meeting. It died because the update part (yesterday, today, blockers) turned into a five-minute morning tax I was already paying in PRs and Jira tickets. I'd close my IDE, open the standup tool, and re-type the same information into a shorter form.

Three weeks ago I stopped. Now Claude reads my activity, drafts the update, I edit it for thirty seconds, and submit. Same content. None of the typing tax. The unexpected part? My EM said the team's standups got better, not worse.

Here's what we tried, what stuck, and what surprised me.

What we replaced

Our old standup was async-by-policy and screenshot-by-reality. Eight engineers, four timezones, a Slack channel pinned to the top of the workspace, and a 9am Sydney "soft deadline" that meant nothing to the engineer in Berlin who'd just woken up. People wrote the update they remembered, not the update that was true. The 4pm thing where you helped a teammate debug a deploy? Forgotten. The PR you shipped before lunch? Forgotten. The blocker you mentioned in standup three days ago? Still there, mentioned for the third time, with no thread connecting the three appearances.

A 2023 Atlassian survey put manager time on status collection at around 17% of the working week. That number tracks for me. The expensive bit was never the meeting itself, which was already 60 seconds of skimming. The expensive bit was the writing and the chasing and the synthesising into something a stakeholder could read on a Friday.

We didn't want to replace the standup. We wanted to remove the typing.

The new workflow, end to end

Every engineer on the team runs one saved Claude prompt in the morning. Five steps, none of which involve typing into the standup tool.

The engineer opens Claude Desktop. Or Cursor. Or Claude Code. Pick your client.
They run their saved standup prompt.
Claude reads yesterday's GitHub PRs, Jira transitions, and (if they've connected it) calendar.
Claude drafts answers for each standup question. The engineer reads, edits anything wrong, adds a real blocker if there is one.
The engineer says "submit". Claude calls Kollabe's standup_submit_answers tool. The update lands in the standup view exactly as if the engineer had typed it.

The whole loop is about thirty seconds on a normal day. On a day with a real blocker it's two minutes, because the engineer adds context the AI can't infer.

The prompt itself is the only thing worth copying:

You are drafting my Kollabe standup for today.

1. Use the Kollabe MCP to find my standup.
2. Pull my activity from yesterday using the GitHub MCP (PRs opened/merged/reviewed,
   commits on branches I own) and the Jira MCP (issues I transitioned, commented on,
   or that were assigned to me).
3. Use Kollabe to get the question list for my standup.
4. Draft an answer for each question, in plain language, no bullets longer than 12 words.
   - "Yesterday" = what I shipped or moved.
   - "Today" = what's actually on my calendar / picked up, not aspirational.
   - "Blockers" = empty unless I genuinely have one.
5. Show me the draft. Wait for me to approve.
6. On approval, submit via Kollabe.

Save it as a Claude Project, a slash command, or a snippet in your client of choice. Mine lives as a saved prompt called /standup in Claude Code so I can fire it from the terminal.

Why this beats a Slackbot, a form, or just yelling in chat

I've used the Slackbots. I've built the forms. I've yelled in chat. None of those solved the real problem, which is that the update was always a half-remembered version of the day. An AI draft built from actual activity is grounded. It catches the 4pm thing because the 4pm thing exists in your commit history.

There's a more boring reason this matters too: structured data. When the standup is text in a Slack channel, the AI summary on Friday is doing pattern recognition on noise. When the standup is structured submissions in Kollabe, the same summary works against typed fields it actually understands. You can ask "which blockers appeared more than once this sprint" and get a real answer instead of a guess.

The other thing took me a sprint to notice: acting identity. The MCP token submits the standup as the engineer, with their role and permissions, against a real audit trail. There's no bot user posting on someone's behalf. The blocker is owned by the person who hit it, and the threading goes to the person who can fix it. Sounds like a small detail. It isn't.

The manager-side win

Before this, my EM read about twelve standups a day, mostly skimmed, asked clarifying questions in DMs. By Friday she'd compile a sprint roll-up by hand, and her Friday morning was that.

Now she runs one prompt:

Summarise the team's Kollabe standups for the last 5 working days
from the "Web" space. Group recurring themes. Flag any blocker that
appears in more than one submission.

Behind the scenes that's standup_list_submissions for the date range, standup_get_summary for the days where Kollabe has already produced one, and Claude clustering across the rest. The output is a markdown digest she can paste into a doc.

For end-of-sprint reporting she runs the same prompt with a fortnightly window. End-of-quarter, monthly, same prompt with a different range. No typing.

She told me last week, almost as a side note: "I used to spend Friday mornings making the report. Now I spend it asking better questions." That's the line that convinced me this wasn't a productivity-hack post. It's a job-shape-changing post.

What about the live standup?

Some teams still gather. Timezone-overlap teams, junior-heavy teams, teams who actually like seeing each other. We used to be one of them, and one of our smaller teams still is.

Live standup with pre-submitted updates is a different meeting than live standup without. Everyone walks in with their answers already in the tool. Nobody reads aloud. The fifteen minutes goes to discussion of blockers and dependencies, not roll call.

Stop reading what's on the screen. The screen already says what you did. The room is for what you need.

That reframe, more than the AI bit, is what changed the meeting for the smaller team. AI just made it cheap to always have the screen pre-filled.

The book-of-record argument

This is the part nobody talks about until they've been on Kollabe for a quarter and then tried to leave.

Slack standups vanish. They scroll, they're searchable in theory, they're un-readable in practice after a week. The blocker mentioned three days ago lives in someone's memory or it doesn't exist at all.

A persistent standup tool that AI-summarises is a book of record for the team's day-to-day. When someone asks "when did we first notice the OAuth issue", the answer is two clicks, not a Slack archaeology session. Sprint review prep becomes one prompt. Onboarding a new manager becomes "go read the last four weeks". Performance review season becomes a thing the data already exists for.

You don't notice you need this until you've had it for a while and then can't go back.

What didn't work

I'm not going to pretend any of this was clean.

For the first two days, Claude over-quoted commit messages verbatim. The standup looked like a git log. Fix: I added "summarise, don't transcribe" to the prompt and the issue disappeared.

One engineer turned off the auto-pull workflow because she likes writing her standup with a coffee. That's fine. Both workflows coexist. Kollabe doesn't care whether the submission came from a chat panel or a keyboard.

The worst one: an engineer connected his calendar MCP to the workflow and Claude pulled the title of a 1:1 ("promotion conversation: ") into the "Today" line. He noticed before submitting. We now strip 1:1 titles from the calendar prompt by default. If you connect calendar to anything that posts, do this first, not after.

These are the kinds of things you'll hit. None of them are dealbreakers. All of them are worth knowing about before you ship the workflow to your team.

Setup, in four minutes

Connect Claude Desktop (or your client) to Kollabe MCP. There's a 60-second setup guide. The config snippet is:

{
  "mcpServers": {
    "kollabe": {
      "url": "https://kollabe.com/api/mcp",
      "transport": "http"
    }
  }
}

Restart the client, approve the OAuth prompt, pick your org. Drop the prompt above into a Claude Project or save it as a slash command.

First standup with the new workflow: review carefully, edit liberally, submit. Second standup: review less. Third: tab-tab-submit and you're done before your coffee's cold.

The point isn't the time saved

It's that the standup is now a real reflection of yesterday, instead of whatever I remembered at 9am.

Our team isn't in the meeting business. We're in the shipping business. Less plumbing, more shipping.

If you want to try it, the Kollabe MCP server is free on the trial. One OAuth click and you're in.

Stop Asking AI to Write Your User Stories (Do This Instead)

Kelly Lewandowski — Thu, 09 Apr 2026 21:10:29 +0000

Most teams using AI in sprint refinement start in the wrong place. They ask it to draft user stories from scratch, then spend the rest of refinement fixing what it got wrong.

There's a better approach, and it doesn't involve handing your backlog over to ChatGPT.

The problem with AI-drafted stories

AI-generated user stories have a specific failure mode: they sound right. Grammatically clean, properly formatted, structurally valid. "As a user, I want to filter results so I can find what I need." That's technically a user story. It could also describe literally any product ever built.

The stories pass a quick glance in refinement because nobody pushes back on something that reads well. Then two days into the sprint, the developer implementing it has five clarifying questions and zero answers.

I've watched this happen. The team saves 10 minutes in refinement and loses two hours in back-and-forth later that week.

Where AI actually helps

The real time savings come from using AI after a human writes the first draft. Specifically:

Expanding acceptance criteria. You write the happy path, then feed it to an LLM and ask: "What edge cases am I missing? What assumptions am I making?" It'll catch empty states, permission boundaries, concurrency problems, and error paths you didn't think about. A Capgemini survey from 2024 found that AI-expanded acceptance criteria reduced rework tickets by about 15%. The time saved in refinement is nice, but fewer mid-sprint surprises is the real win.

Catching dependencies. If you give the model your data model or API surface alongside the story, it's surprisingly good at flagging cross-team dependencies and migration risks that slip past human review. The trick is context. A prompt with just the story text gives you generic output. A prompt with the story plus your schema gives you specific flags you can act on.

Splitting big stories. When a story is clearly too big for a sprint, prompting an LLM to "split by user workflow step" or "split by data variation" works better than asking for a generic breakdown. The pattern you give it matters more than the model you use.

A workflow that works

Here's the pattern we've seen work well for teams:

Product owner writes rough draft stories with basic acceptance criteria (doesn't need to be polished)
Feed each story to an LLM with relevant context (data model, related stories, constraints) and ask it to list edge cases, implicit assumptions, and missing scenarios
Review the AI output as a team during refinement. Discard the noise, keep the genuine catches
Estimate with the fuller picture. Stories that went through this process tend to surface complexity earlier, so you get fewer "wait, what about..." interruptions during planning poker

Some teams report refinement sessions running 20-30% shorter. But the bigger payoff shows up later in the sprint when clarification requests drop.

The stuff to watch out for

False completeness. The AI generates 12 acceptance criteria and the team assumes it's exhaustive. It's not. The model can't know what it doesn't know about your system.

Skill erosion. If your junior devs stop learning to break down work because AI does it for them, you've traded short-term speed for a long-term problem. Have less experienced people write the first draft, then use AI to expand on it.

Generic prompts, generic output. "Write a user story for search" gives you nothing useful. "Write a user story for full-text search across project names and descriptions, for a user managing 50+ projects" gives you something you can actually work with.

When to skip it entirely

Bug fixes with clear repro steps, copy changes, and straightforward CRUD work are all faster to refine the old way. Save the AI pass for stories where the problem space is fuzzy or your team keeps discovering unknowns mid-sprint.

I wrote a longer piece on this with more prompting examples and a section on specific pitfalls. If you want the full version, it's here: AI-Assisted Backlog Refinement: Using LLMs to Write Better User Stories

Your Agile Ceremonies Weren't Designed for 10 Time Zones

Kelly Lewandowski — Tue, 10 Mar 2026 20:01:39 +0000

The Scrum Guide doesn't mention time zones. It was written for teams that could stand in a circle every morning, hash out sprint scope over a whiteboard, and grab coffee together between meetings.

That's not most teams anymore. If your engineers sit in New York, Berlin, and Bangalore, you're dealing with a 10.5-hour spread. Sprint planning at 9am Eastern is 7:30pm in India. Your "quick retro" at 4pm Berlin time hits Bangalore at 8:30pm.

The usual fix is rotating who gets the bad meeting time. That's fair, but it still treats every ceremony as a synchronous event. And that's the actual problem.

Not every ceremony needs a meeting

This is the question most teams skip: which ceremonies actually require everyone talking at the same time?

Ceremony	Needs sync?	Why
Daily standup	No	Written updates are faster to consume and don't require timezone coordination
Sprint planning	Partially	Scope negotiation needs real-time discussion, but context-sharing doesn't
Sprint review	Yes	Live stakeholder feedback is the whole point
Retrospective	Yes	Honest team conversations about dynamics need tone of voice and real-time energy
Backlog refinement	Hybrid	Async pre-read, sync for questions and estimation

The daily standup is the easiest ceremony to move async, and it frees up your overlap window for the ceremonies that actually benefit from live discussion.

The overlap window

Map out when your team members are all within working hours. For most globally distributed teams, this is 2-4 hours. Some get less.

For that NYC/Berlin/Bangalore spread, the overlap is roughly 14:00-16:00 UTC. Two hours. That's it.

Those two hours are sacred. One meeting per day, max. Everything else happens async. If you're burning your overlap window on status updates, you won't have time left for sprint planning or retros, which are the ceremonies that actually suffer without real-time discussion.

A small trick that helps: ask people to flex 30-60 minutes in either direction. A Berlin dev starting at 10am and a Bangalore dev staying until 7:30pm buys you an extra hour. Rotate who flexes so nobody's always the one adjusting.

Sprint planning without the 2-hour meeting

The async-prep-sync-decision pattern cuts planning meetings in half:

48 hours before: PO shares candidate backlog items with acceptance criteria and context. Team reads and posts questions async.
24 hours before: Team runs async estimation (planning poker works well here — everyone votes independently and you can spot disagreements before the call).
During overlap: Live session focuses only on resolving disagreements and committing to the sprint goal. 45-60 minutes instead of 2 hours.

The information transfer happens async. The negotiation happens sync. You stop wasting synchronous time on things people could have read on their own.

Why retros need to stay synchronous

I'd argue retros are the ceremony you should fight hardest to keep live. The candid, sometimes uncomfortable conversations about how the team works don't land the same way in a shared doc. Text strips out tone. Written responses feel less safe than spoken ones.

That said, you can make the sync portion shorter:

Have people add retro items to the board before the meeting. This gives everyone, especially those in less convenient time zones, equal chance to contribute.
Keep it to 60 minutes. Distributed retros lose energy faster than in-person ones.
Use anonymous voting. Power dynamics get amplified on screen.

Distributed teams need retros more than co-located ones. Miscommunication and unclear handoffs pile up silently when there are no hallway conversations to catch them. The retro is where that stuff surfaces. Skip it and the problems just compound.

The sample week

Here's what it looks like in practice with a 2-hour overlap (14:00-16:00 UTC):

Day	Overlap window	Async
Monday	Sprint planning (60 min)	Standup updates, planning pre-read
Tuesday	Open for ad-hoc sync	Standup updates, refinement pre-read
Wednesday	Refinement (45 min)	Standup updates, estimation
Thursday	Open for ad-hoc sync	Standup updates
Friday	Retro or review (60 min)	Standup updates, retro board input

One meeting per day in the overlap window. The rest of the time, people build things.

What usually goes wrong

Defaulting to HQ time. If leadership is in New York and every meeting happens during East Coast hours, your other offices are permanently accommodating. People notice and stop engaging.

Skipping documentation. Co-located teams have hallway conversations. Distributed teams don't. If you didn't write it down, it didn't happen for anyone who wasn't on the call.

Adding more meetings to compensate. The instinct when async communication feels lacking is to schedule more syncs. This makes it worse. Fix the async communication instead.

I wrote a longer version of this with concrete team agreements, sprint review strategies for multiple timezone clusters, and a deeper breakdown of the async-first approach. You can read the full post on the Kollabe blog.

If your team is already doing async standups or hybrid planning, I'd love to hear what's working. Drop a comment.

Your team ships faster with AI. Here's why you need retros more than ever 🤖

Kelly Lewandowski — Sun, 01 Mar 2026 19:27:13 +0000

70% of developers say AI coding tools make them more productive. Only 17% say those tools improve team collaboration.

That stat from Stack Overflow's 2025 survey stuck with me. We're all shipping faster individually, but nobody's talking about the team-level side effects.

The numbers that should make you uncomfortable

A few data points that changed how I think about this:

METR study: Experienced devs using AI tools took 19% longer to complete tasks, while believing they were 20% faster. A 40-point perception gap.
DORA 2024: A 25% increase in AI usage correlates with a 7.2% decrease in delivery stability.
GitClear: Code churn jumped from 3.1% to 7.9% between 2020-2024. Refactoring dropped from 25% to under 10%.

More code is shipping. Whether it's the right code is a different question.

The collaboration problem nobody's fixing

A two-year longitudinal study found that AI adoption shifts work toward individualized coding tasks and away from collaborative coordination. The collaboration problems that existed before AI (silos, communication gaps, unclear ownership) stayed completely unresolved.

AI is making individuals faster. It is not making teams better.

Where retros come in

When your team spends 90% of the sprint heads-down with an AI pair programmer, the 60 minutes in a retro might be the most important hour in the entire sprint.

The questions need updating though. For AI-assisted teams, these are the ones that matter:

AI tool effectiveness - Where did AI help? Where did it waste your time?
Knowledge distribution - Who actually understands the code that shipped?
Customer connection - Did shipping faster translate into customer value?
Code quality signals - Is churn going up? Are PRs getting rubber-stamped?
Team norms - What are your unspoken rules about AI usage?

The DORA quote that sums it up

"AI makes good teams great. And bad teams worse, faster."
-- Google DORA 2025 Report

The practices that separate good teams from bad ones (psychological safety, shared understanding, honest feedback) are exactly what retros are built around. As AI handles more of the mechanical work, the human conversations get rarer. And rarer means more valuable.

I wrote a longer piece diving into the research and practical retro questions for AI-assisted teams: Read the full post on Kollabe

User Personas That Developers Actually Care About ⛄️

Kelly Lewandowski — Wed, 25 Feb 2026 18:27:48 +0000

Most user personas end up in a slide deck that nobody opens after sprint two. They read like a casting call for a fictional character, packed with hobbies and favorite coffee orders but missing anything that would change how your team builds software.

I've been thinking about why that is, and I think the problem is that most personas are written for product managers and designers, not for the full team. Developers never see them, so they never use them. And a persona nobody uses is just creative writing.

Here's what actually works.

Keep it to one page

A persona that fits on a single page gets used. One that requires scrolling gets ignored. Here's what belongs:

Section	What to include
Demographics	Name, job title, company size, location
Goals	3-5 professional goals related to your product
Pain points	Current frustrations and blockers
Behavioral patterns	How they work, buy, and make decisions
Technical context	Devices, tools, comfort level with tech

What to leave out: favorite color, what they eat for breakfast, detailed personal backstories. If a data point wouldn't change a product decision, it doesn't belong.

A quick example

Here's a persona for a B2B project management tool:

Sarah Chen, Senior Engineering Manager, Series B SaaS company (120 people)

Goals: Ship features predictably, reduce meeting overhead, give leadership visibility into sprint progress without daily check-ins.

Pain points: Current tools don't surface blockers early enough. Sprint retros feel repetitive. Estimations are inconsistent across sub-teams.

Behavioral patterns: Prefers async communication. Reviews dashboards every morning before standup. Evaluates new tools by trying the free tier herself before involving procurement.

"I don't need another dashboard. I need my team to spend less time talking about work and more time doing it."

Every detail here connects to a product decision. Sarah's preference for async work means your product needs strong notifications and reporting. Her evaluation behavior tells you the free tier has to be self-serve and compelling.

Where personas actually get used

A persona that doesn't show up in daily work is just a character sketch. Here's where they actually get used:

During backlog refinement, ask "Which persona does this serve?" If the answer is "all of them," the story is probably too broad. If the answer is "none of them," it might not belong in the backlog.

When writing user stories, replace "As a user" with the persona's name. "As Sarah, a remote engineering manager, I want to see sprint progress without attending standup" is immediately clearer and easier to estimate.

In sprint planning, if your primary persona's top pain point isn't addressed in the upcoming sprint, that's worth flagging.

And in retros, try asking "Are we building for our personas, or have we drifted?" Two-minute gut check, surprisingly useful.

How many do you need?

Three to four. More than that and your team can't keep them straight. Fewer than two and you're probably treating a diverse user base as a monolith.

Cover these types:

Primary user - the person who uses your product daily
Buyer - the person who makes the purchase decision (often different in B2B)
Influencer - someone who recommends or blocks adoption

You might also create a detractor persona to understand who your product is not for. Knowing who to say no to is just as useful as knowing who to say yes to.

Don't wait for perfect research

Start with provisional personas based on what you know today. Pull from support tickets, sales call notes, analytics. Five to eight user interviews per segment is enough to spot patterns.

A rough persona that the team actually references beats a polished one that nobody reads. You can always refine later.

The 5-minute version

If you want to get a structured persona fast, I built a free AI User Persona Generator that creates one from a product description in seconds. It's not a replacement for real research, but it gives you something concrete to react to instead of starting from a blank page.

I wrote a more detailed version of this guide with step-by-step instructions and more examples on our blog: How to Create User Personas That Actually Improve Your Product

If your team uses planning poker or retrospectives, you might also find these useful: