You can finish an entire interview loop and then find out you need more interviews because you were interviewing for the wrong level. I've seen it multiple times. People get all the way to the end, realize they were in the wrong loop, and then either do more interviews or stay at the lower level. Before round one, ask your recruiter "Which level am I being considered for?" and decide whether you'd actually take it. Sometimes you can get upgraded during the process, but just know that it might not happen, and the loop might be a waste of time. If you really like the company, it might be worth taking that risk and trying to argue and convince them to up-level you.

"My problem with AI is that it now excels at the part of the job that was my favorite (writing code) and it leaves me to do the part that I suck at (being a manager)." That's from a Hacker News comment that's going pretty viral. The engineer, who goes by ryandvm, added that in 18 months or so his job satisfaction has gone from "pretty high to abysmal." Not everyone in the replies agreed. Plenty of engineers said they're enjoying building more now, but many others said they feel the same way he does. It's almost the opposite of what I'm hearing in product. I'm involved in Supra, a community of product leaders, so I talk to them all the time, and a lot of them are having a ton of fun right now. I keep meeting leaders who used to run big teams of PMs and are now senior ICs with zero reports, and they're enjoying the work more than ever. They're falling back in love with the craft because they can build things themselves again. I've been trying to figure out why the same tools feel so different depending on which side of the table you sit on. My hypothesis is that what we call product craft was never really about producing artifacts. Most of a PM's job was about deeply understanding the user problems, prioritizing the most important opportunity, and aligning people to build the best solution. And many of us wished we could build the thing ourselves. So getting way more agency over the output with AI feels like a gain. For a lot of engineers, writing the code was the craft, and that's the part AI is taking over. And both groups now spend a big part of their day directing agents. PMs come to it from aligning a room full of people, which is way harder, so it feels like a relief. Engineers come to it from writing code, the part a lot of them loved most, so it feels like they lost the craft and got handed management instead. Their job changed under them, and I wonder how many companies have actually sat down with their engineers and clearly told them the job has changed, the craft has changed, here's what we expect from you now. I'm not sure that reset has happened in a lot of places. So if you work with engineers, check in with them and see how they're doing with all this. Maybe even give them a hug.

The median Anthropic employee spends 20% more time in meetings than a year ago. Even technical individual contributors spend 16% more time in meetings. Anthropic published a report last week on how work has changed inside the company. They put it together from their own data: hiring, calendars and Slack. It's no secret Anthropic has grown a lot, and that's probably a big part of the spike in meetings. But I think another part is that as you create more and move faster, getting aligned becomes even more important. Many of the other numbers point the same way: ↳ Interactive Claude sessions spend around 80% of their time sitting idle, waiting for a human to respond. One engineer put it as people "bottlenecking the agents." ↳ Meta-work, which is basically setting up and keeping an eye on agents instead of doing the task yourself, went from 10% of the average employee's Claude sessions in fall 2025 to 22% in spring 2026. In an April survey, 20% of employees said their role shifted from doing work to managing Claudes, and one engineer said it feels like facing "an army of Claudes, with every one asking for my attention." ↳ There are around 17 engineers for every PM now, and that ratio's gone up since 2025. The share of people on technical teams who aren't actively coding fell by almost half, and commits from managers doubled. ↳ On hiring, juniors went from 27% of technical hires to 9%, and junior roles are just 6% of technical openings now. Around a third of ML managers and a fifth of software engineering managers said they're hiring less because of AI. ↳ Agents show up in nearly half of active public Slack channels. The report calls private work in DMs and unlogged meetings "invisible, confusing and discouraged," and one researcher said "leaders have to defend feedback out in the open." One non-technical employee said "that 20% that was strategic becomes 90% of the job." My read is that AI made a lot of the work cheap. What's scarce now is people's attention and judgment, plus the trust you still build face to face in all those extra meetings, which is basically the job leaders already had. Somebody has to decide which agent output gets a human read, because nobody can read all of it. Somebody has to figure out where judgment comes from when way fewer junior employees get hired, and even the authors admit none of their fixes "directly address the slowdown in junior hires." And since agents only know about decisions that get written down, feedback that used to live in a DM needs a more public home, which I'd guess many leaders find uncomfortable. And to be clear, Anthropic is not a normal company. They build frontier models and use them daily; the company tripled in size in the last year, and the authors say many of its employment practices can't be replicated across the economy. Still, it's the clearest look I've seen at what happens when agents do a lot of the work. And most of what got harder is the people stuff.

If your behavioral round felt like a checklist with no follow-ups, you probably got an interviewer who's too junior or hasn't done many of these. Say they ask "Tell me about a time you disagreed with a teammate." You give a solid answer, they nod, and they move straight to the next question. No "why," no "what happened next." You walk out feeling like you never got to show what you've actually done. We talked about exactly this with Austen McDonald, who chaired engineering hiring committees at Meta, on the latest episode of Supra Insider, the podcast I co-host with Ben Erez. He says less experienced interviewers tend to read down a list and don't ask many follow-ups. So steer it yourself. Before you answer the narrow question, give some context on the project, like the stakeholders you had to align or the hard problem you solved. And if it still stays shallow, I'd tell the recruiter. Something like "I feel like this conversation didn't go super deep, and I didn't get a chance to share X." On Austen's committees, they looked for chances to fill gaps like that, and recruiters will often try to find someone else for a deeper follow-up. I'd lead with what you didn't get to cover and mention the interviewer second. My guess is people appreciate that kind of transparency, but leading with the interviewer can sound like "I'm too good for this." It might not get you a pass, but as Ben put it, it gets you your day in court. Btw, if you've got behavioral rounds coming up, the whole episode is worth a listen.

An L7 principal PM at Amazon started presenting a six-page memo, and two minutes in, everyone in the room could tell he was reading it for the first time. Steve Huynh shared that story from someone in his community who was in the room. Nobody said anything, because he was the most senior person there. Huynh's point is that a six-pager used to be proof the author had done the thinking, and with AI that's just not true anymore. Clay, the go-to-market data platform, has an AI writing policy that gets at the same problem. Sophie Alpert wrote it for their engineering team, and other teams liked it so much that it became company-wide. The first rule is "You must stand behind every idea and every sentence in your docs." So if a reviewer asks what you meant by a line, "Oh sorry, AI wrote that, just ignore it" doesn't count as an answer. Without a policy like this, I think trust is the first thing to go. Once people suspect the author hasn't really lived what's in a doc, they start skimming, and once nobody's reading closely, there's even less reason to think things through before writing, so the bar keeps dropping. I'm not going to get on a high horse about this. I've definitely shared AI-created docs I hadn't fully read, couldn't stand behind every line, and I'm not proud of it. At Insider Loops, Ben Erez and I are trying to hold ourselves to a similar rule. If one of us can't answer for every line, whether it's in a guide or a message to each other, it doesn't go out. We call those slop grenades, and sending one isn't fair, because if you haven't done the thinking, why should the other person take the time to carefully review what AI wrote?

I used to picture the sleazy car dealership guy whenever I thought about sales. Now I mostly think of it as helping someone with their problem and figuring out if there's a match. I hosted a Supra session with four product leaders who left their jobs to go out on their own, and a lot of what they wish they'd known was about selling and distribution, way more than the product itself. Here were some of my favorite nuggets: ↳ One spent way too much time on the customer problem "because I'm a product manager." If you can't prove how you'll reach customers, don't quit yet. ↳ Another built an app, put it out, and nobody saw it, all in about a month. Building isn't the valuable part anymore. ↳ Sit in on sales calls while you still have a salary, as product discovery. The same one says selling ended up harder than the delivery work. ↳ Be shameless about asking. One said nobody's counting how many times you post the same ask. You're going to suck at selling at first, so get comfortable being bad and keep doing it until you're not.

I’m fine with an agent making a few assumptions as it figures things out. I’d just like it to check with me on the ones that could turn into an afternoon of “no, that’s not what I meant.” On Latent Space, Anthropic’s Thariq Shihipar talked about Claude Code’s Ask User Question tool and how it can help the agent draw out what someone wants. As more people use Claude Code, choosing when it should ask gets harder. Some people feel they’ve already given it enough detail and want it to get on with the work. Others need help figuring out the requirements and their own preferences. He thinks almost everyone has more to figure out than they realize, including decisions about how to structure the data that are better settled before it starts building. I use AI tools all day, every day, and most of the time I’m still not totally clear on what I want. The impatient side of me wants the agent to get cooking asap, but I’m also relying on it to help me figure things out as we go. I’m starting to pay more attention to how much time I spend correcting the agent or having to start over because we went down the wrong rabbit hole. At this point, answering a couple of questions at the beginning feels a lot less annoying.

If your interview has a take-home presentation, make it look like you already work there. It's easy, and very few people do it. Ben Erez and I recently talked to a PM who got his offer after a take-home presentation. Before he built his deck, he found a copy of the company's design system that someone had posted online, which is basically the fonts, colors, spacing and icons their own teams use, and had Claude build his slides in it. Now that he works there, he says his slides look just like the internal ones. He also showed the deck to a friend who had sat in a lot of executive reviews, and the main thing his friend told him was to move his recommendation up to slide 2, right after the intro. When he presented it that way, he said the whole room sat up, because they wanted to know why he was recommending it. I talk to a lot of people doing take-homes, and I rarely see this. Before you start on the slides for your next one, search for the company's design system online. If the take-home rules allow AI tools, Claude or something similar can apply it to your deck. If they don't, you can still match it by hand or look for the company's PowerPoint templates. Then give the deck to a friend who hasn't seen it, and ask for more than feedback on the content. Ask whether the structure and the high-level points land, and whether you're really starting with a point of view, ideally with your recommendation right at the top. This detail alone won't get you the job, but if you compound enough of these little details that show that you care and that you are different, that will make a difference in the long run.

Searching for your next product leadership role but not sure where to look? Every two weeks, I'll be posting a curated list of open Product Leadership Roles at leading companies to help people who are actively looking for jobs. For context: I run Supra, a private community of high-caliber product leaders, so I have access to roles our community members are hiring for and jobs posted in our job-openings Slack channel. Here are the most interesting roles: Prefer to get the the roles via email 📧? Subscribe to this bi-weekly job listing here: buff.ly/nbHOLt8. 𝐇𝐞𝐚𝐝 / 𝐕𝐏 𝐨𝐟 𝐏𝐫𝐨𝐝𝐮𝐜𝐭 𝐫𝐨𝐥𝐞𝐬 Mapbox is hiring a VP of Product buff.ly/NwpYxwQ Intentsify is hiring a Vice President, Product Management buff.ly/x1670Vi SpotOn is hiring a VP of Product Management buff.ly/OnGCimZ BrightHire is hiring a Head of Product buff.ly/ThYOkmp Rapid7 is hiring a Vice President, Product Management (MDR) buff.ly/JhpLuuu Formlabs is hiring a Head of Product SLS buff.ly/Y6RglU5 General Medicine is hiring a VP Product buff.ly/HXmxHot 𝐃𝐢𝐫𝐞𝐜𝐭𝐨𝐫/ 𝐒𝐫 𝐃𝐢𝐫𝐞𝐜𝐭𝐨𝐫 𝐨𝐟 𝐏𝐫𝐨𝐝𝐮𝐜𝐭 NVIDIA is hiring a Director, Product Management Nemotron buff.ly/PMQTG59 1Password is hiring a Director of Product, AI Growth buff.ly/SSjteSo Verkada is hiring a Director of Business Systems Product Management buff.ly/PHLkioH Moloco is hiring a Director of Product Management, Ads Quality and Performance Moloco Commerce Media buff.ly/1g28ZWf StubHub is hiring a Director of Product Marketplace Policy buff.ly/f3zTBRW Meta is hiring a Director, Product Management Agents and Intelligence buff.ly/R6RiZ6I Google is hiring a Director, Product Management, Cloud Networking buff.ly/XvwOFYt Ethos is hiring a Sr. Director, Product Management, Consumer buff.ly/Q4zAM0C Uber is hiring a Director, Product Management Agentic AI jobs.uber.com/en/jobs/301366… 𝐆𝐫𝐨𝐮𝐩 𝐏𝐌 / 𝐏𝐫𝐨𝐝𝐮𝐜𝐭 𝐋𝐞𝐚𝐝 Google is hiring a Group Product Manager, Frontier Model Protection, DeepMind buff.ly/Rr1cRdq Coinbase is hiring a Group Product Manager, Customer Engagement & Experience buff.ly/ZCwSpXO Procore is hiring a Group Product Manager Labor and Equipment buff.ly/M3rr051 Uber is hiring a Group Product Manager, Grocery & Retail Advertising buff.ly/gKn0Prw 1Password is hiring a Senior Group Product Manager (Apono) jobs.ashbyhq.com/1password/0… Juniper Square is hiring a Group Product Manager, AI-Native Data Warehouse buff.ly/9x1gRkD 𝐏𝐫𝐢𝐧𝐜𝐢𝐩𝐚𝐥 / 𝐋𝐞𝐚𝐝 𝐏𝐌 Meta is hiring a Product Manager Lead, Integrity Measurement buff.ly/Wg3pOD9 Amazon is hiring a Principal Product Manager Technical, Platform Engineering & Emerging Technologies (AWS) buff.ly/ss4uk4u Microsoft is hiring a Principal Product Manager CoreAI (Enterprise Security and Governance) buff.ly/9EAk88Z Uber is hiring a Lead Product Manager Earner Growth Experience uber.com/careers/list/157504 Writer is hiring a Lead product manager jobs.ashbyhq.com/writer/ceb3… OpenAI is hiring a Product Manager, API Agents openai.com/careers/product-m… Anthropic is hiring a Product Manager, Safeguards Rare Harms job-boards.greenhouse.io/ant… Spotify is hiring a Senior Product Manager, Consumer Audiobooks jobs.lever.co/spotify/c5c82f… Citian is hiring a Product Manager job-boards.greenhouse.io/cit… OpenAI is hiring a Product Manager, API Infrastructure openai.com/careers/product-m… Google is hiring a Senior Product Manager, Discover Recommendations google.com/about/careers/app… 𝐓𝐨 𝐬𝐞𝐞 𝐦𝐨𝐫𝐞 𝐫𝐨𝐥𝐞𝐬, 𝐜𝐡𝐞𝐜𝐤 𝐨𝐮𝐭 𝐨𝐮𝐫 𝐒𝐮𝐩𝐫𝐚 𝐣𝐨𝐛𝐬 portal 𝐰𝐢𝐭𝐡 +80 𝐨𝐩𝐞𝐧 𝐫𝐨𝐥𝐞𝐬 > coda.io/@supra/supra-job-boa… If you know someone looking for a job or hiring, tag them in the comments or repost this post to help spread the word. Follow me (Marc Baselga) to receive these job openings.

If a recruiter asks what salary you’d accept before you even understand the role, have an answer ready. That’s a pretty expensive thing to wing. Colin LernellI talked about this when Ben and I had him on the podcast. Even experienced product leaders get worn down by the question and eventually give a number because they just want to keep things moving. I’d have something ready along these lines: “I’d like to understand the scope and level a bit better before putting a number on it. What range have you budgeted for the role?” Then think through what you’ll say if they ask again. Colin sees this come up several times during a process, including when someone’s busy prepping for the next interview and tempted to just throw the recruiter a number. The recruiter can leave that conversation thinking that’s what it would take for you to accept. If you learn more about the job and decide you’d need more, you now have to explain why your number changed. You can still negotiate, but you’ve given yourself extra work before you even have an offer. And Colin was clear about the exception too. If they genuinely need an answer to move you forward, getting the offer still matters. I wouldn’t lose an opportunity just to avoid saying a number. Before your next recruiter call, spend a few minutes practicing your answer to the compensation question, including what you’ll say if they push. You’re already putting hours into interview prep. This deserves a little of that time. Ben and I get into this with Colin Lernell on the latest episode of Supra Insider, including what to say when the recruiter keeps pushing for a number. If you’ve got interviews coming up, give it a listen.

Most PM loops we've looked at this year never ask how you use AI. They let the take-home answer it. When there's a take-home, AI fluency shows up in the depth of the submission. People who get a lot of leverage from AI use the same window to go way deeper, with more evidence behind each call and a far more detailed prototype if one's asked for. A beautiful deck and polished writing don't prove much anymore, since AI gives everyone those, and on their own they can read as slop. If your take-home allows AI, I'd spend it on depth over polish. Pick the few calls that matter and go further on those instead of covering everything. Use AI to pull in more evidence behind them, and to build a prototype that actually works, even a rough one. Say up front how you used AI, and keep the thinking yours, because if they ask about your calls, it shows pretty quickly how deep you really went.

Some news I’m super excited to share: Joshua Herzig-Marx is stepping in as General Manager of Supra and leading the community day-to-day. Being able to bring someone I respect and admire so much into this role feels pretty surreal. I’ll still be involved in a different way, focusing more on Supra’s direction and programming and finding ways to support our members, while Josh leads the team and runs day-to-day operations. Supra means a lot to me, and I’m excited to keep building it together. People often ask me why I started Supra. I always wanted to be an entrepreneur, but I’ve also always been risk-averse. The way I rationalized starting Supra was: best case, I build a business I’m proud of. Worst case, I spend time with people I respect, learn a ton from them, and build great relationships. I was very comfortable with that worst case, which pushed me to actually start. Almost four years later, that part of building Supra has wildly exceeded my expectations. I’ve made close friends, met business partners, and gotten to know people I’d looked up to for years. Through Supra, I’ve helped people meet their co-founders, find fantastic teammates, and work through important moments in their careers. Being part of all of that has been one of the most rewarding experiences of my career. Josh was one of the first coaches who took a chance on me, and he’s been a trusted voice ever since. Having him help lead something we both care so much about feels like a beautiful evolution. He brings experience and strengths I don’t have, while sharing the belief that made me want to build Supra in the first place: the quality of the people and their relationships come first. I’m sharing the note I sent our members below, with more about Josh and what this means for the community. I’m more bullish on Supra than ever. With so much changing in tech and product leadership, having people you trust to learn alongside and talk openly with feels more important than ever. Today I’m mostly feeling a lot of gratitude. For my family, for the people who believed in me when I didn’t, and especially for the early members who took a chance on a guy with a lot of enthusiasm and a half-baked vision. I’ve poured everything I’ve got into Supra, and I’m so excited to see what we can build with Josh leading the team. If you’re a product management or product design leader and have been thinking about joining, message Josh or me. We’d love to tell you more about what we are cooking. The best of Supra is still ahead of us 🚀

So many people ask me for intros and then send me mediocre blurbs. If your blurb doesn't give me enough confidence that you understand the role and how you fit it, I won't send that intro. Not worth risking my social capital for an intro that I have low confidence in. Some quick dos and don'ts. ↳ Do write it to the hiring manager, not to me, so I can forward it without changing a word ↳ Do open with the problem the role exists to solve, then why you're the person to solve it ↳ Don't ask me for a crazy intro if I don't know you, have someone I know and trust vouch for you first ↳ Don't count on the big names on your resume to do the work ↳ Don't stretch a random project from five years ago into a "perfect fit" Yes, that's maybe 15 or 20 extra minutes of work, but it's worth it. Getting in the door through a cold connection or application is way harder. Use your time wisely, friends.

Reminder that you're probably not as good at negotiating as you think you are. I talk to a lot of people who tell me "I don't need help with the negotiation, I'm already really comfortable pushing back and I've read {insert_fav_salary_negotiation_book}." But the more I learn about negotiating in tech, the less I think people actually know. There's just so much to it; it's a very complex dance that takes practice and skill. Partnering with someone who knows what they're doing is super valuable. Please don't negotiate alone, since you'll likely leave a lot of $$$ on the table.

If you want to be a CPO someday, there's a good chance people three or four levels below you get a call about you before you get the offer. We had Nolan Church on Supra Insider. He ran talent at DoorDash during hypergrowth, was Chief People Officer at Carta, and now runs exec searches himself, so Ben Erez and I asked him how companies actually decide on an exec hire. Here's what we learned. 1/ Treat everyone you work with like a future reference When Nolan checks references, he talks to basically everybody. Your manager and peers, but also your direct reports, people three or four layers down in your org, and the board. Some of those calls are back channel, so they won't all be people you put on a list. Honestly, I'd just be kind to people, especially the ones a few levels below you who you don't see every day. 2/ Know what your detractors would say At the top, everyone has detractors, and Nolan isn't looking for ten out of ten. What he does do is take the negative stuff back to the candidate ("hey, this is one of the things I heard") to see how self-aware they are. So it's worth thinking now about the criticism you're most likely to hear and how you'd talk about it honestly. 3/ Expect to work on a real problem It depends on the company, but one of the best ways to get signal at this level is having you work through something they're dealing with right now, sometimes as a case study and sometimes in the trenches with the team. Be ready to think through their actual problem with them. 4/ Assume it's already started The casual conversations count too. Nolan picks up a lot of small signals in them, down to how fast you reply to a text. The full conversation with Nolan is on Supra Insider, including why companies still hire execs when everyone's talking about agents, and how he gets people who aren't looking to take the call.

AI agents inside Stripe nearly took down production. Stripe has an internal AI agent called Kai that more than 10,000 employees use every week. Sharadh Krishnamurthyamurthy, the engineering manager who helped build it, went on Claire Voo's How I AI and talked about what happened when agents started behaving unpredictably and overwhelming core systems. His argument is that agents mostly amplify the weak spots a company already has, hitting them way harder and way more often than people do. As he put it, "when in doubt, an agent will just brute force it." A person who can't find a number in a dashboard eventually gives up and asks someone, while an agent will happily run query after query until it gets one. Better prompts weren't enough to solve it, so Stripe went after the plumbing. Agents got their own identities, which let systems tell an agent's request apart from a person's. Stripe also added priority signals and load shedding, which basically means the less important work gets dropped first when things get overloaded, instead of everything going down together. The last piece was a query layer (the layer agents go through to pull data) built to absorb all the extra demand. A lot of this sat on top of work Stripe had already done over the years for its human engineers, like a solid data platform and clean APIs. If you're building or speccing an agent right now, these are the questions I'd want answered before it ships. ↳ Does it act as itself, or borrow a person's access? If it borrows, nothing downstream can tell the two apart. ↳ What happens when it does the same thing 1,000 times in an hour? ↳ When things get busy, whose work gets dropped first, the agent's or a customer's? ↳ Which actions need a human to say yes first? Stripe is hiring a PM to work on a version of these questions for its customers. The Identity & Access Management role, open right now, includes deciding what agents can do on Stripe and when a human has to approve before money moves. Even if you're not going for that role, Stripe's technical round tests similar instincts at the API level. It's run by an engineer, and one candidate described it as the interviewer hunting for "the depth you're uncomfortable at." Be ready to explain why you'd rate-limit a usage-based API, and how you'd check who's calling, like why you'd use a bearer token instead of basic auth. P.S. This is one slice from the Insider Loops Stripe guide. We captured the complete Stripe PM interview process at Insider Loops

Nashville-based product leaders: We’re hosting our first Supra meetup in Nashville next Tuesday! Our member base there has been growing, and I can’t wait for everyone to get together. RSVP here: buff.ly/SOsPrNP Huge thanks to Kelly Farr for leading the charge. She’s been key to helping us grow the community, and I can’t wait for all of you to meet her.

Andrew Somervell's 10-person team at Hamster doesn't do sprint reviews or planning poker anymore. They have three meetings a week, and two happen before anyone commits to building anything. I asked him to walk Ben Erez and me through it on Supra Insider. They used to have meetings to estimate work and figure out where things stood. Now Andrew says he can see where things are in the thread for each piece of work. Those threads live in Hamster, his own product, but if you can already see where the work stands, why have a meeting to find out? They still spend time deciding what to build and making sure everyone's heading in the right direction. Because, as Andrew put it, someone going the wrong way "can go real fast with AI now." Here are the three meetings they have at Hamster: 1/ Direction. Once a week, 45 minutes to an hour, with four people: the co-founders, product and their lead engineer. They look at a goals page with live metrics and talk through what needs to change. Quarterly planning doesn't really work for them anymore. As Andrew put it, "if you can get six to eight weeks out in advance, like good luck." Someone on his team even asked whether engineers should see the goals page. His answer was that everyone should know exactly why they're doing something. 2/ Discovery. This is Andrew's favorite. People bring whatever they've got: a 30-second whiteboard squiggle, a screenshot, a customer insight, a proof-of-concept branch an engineer put together. Then they talk it through. Nobody commits to building anything in that meeting. 3/ Delivery. This is where they commit to the week. They still do design crit at the end of delivery, and Andrew wants the review to come back to the original bet: "what was the thesis? What were we gonna try and improve?" Hamster's write-up of their method also makes the case for keeping scopes small. A brief an agent can finish in thirty minutes beats one that takes three hours. The part I'd steal, even at a big company, is having separate meetings to explore ideas and commit to work. When you do both in the same meeting, I think it's easy to go from "we're just talking about it" to "okay, I guess we're building it." I'd be more careful about cutting status meetings, since those often also cover cross-team dependencies and reporting up. But I'd still ask which ones we're having just because nobody can see where the work is.