3 Weeks to Your First AI PM Role: A No-BS Roadmap | Real Hiring Signals from Big Tech

0:05 I've seen like four or five candidates now  in entropic where everything is perfect just they are saying you are not a cultural fit  because you have not shown enough that why anthropic they will ask you simple questions  like hey why you want to join anthropic and that question means specially something  special anthropic is doing which meta is not doing which openai is not doing which  super intelligence labs are not doing.

0:38 I'm glad I'm here and I can see all of you  are here too. Who am I then? Only to the new people. I'm Mahesh. I've worked at all the big  companies. Uh that was one of the bucket list I had and I finished it. So what I was doing is I  built lot of agents when I was at Google. I was trying to automate their back office function at  GCP. Uh before that I was building frameworks that people can use to build agents at AWS and before  that I was building the large language models so that people can go and b build the technology  which is powering these agents and before that I was working at Microsoft as a developer and then  as a product manager. So my journey is now almost 20 years. I have spent 16 17 years as a developer  and last 5 years or four five years as a product manager. Now I have started my own company where  I'm trying to help bring the benefits of AI to enterprises by showing them what's the impact  of agents to their companies and also building custom agents for them for back office. So  if you have a legal function, HR function or accounting function, we are building agents for  you. So that's my life. The exciting part of my life is that I am also loving to teach. So I'm  teaching for I don't know since my college days but formally I started teaching teals students.  Uh this was a program at Microsoft where I taught high school students how to code. Then I taught at  uh Facebook. I taught recommendation system course and then I taught this course which you can look  at Maven for last 3 years and I have done almost uh 15 cohorts. uh we didn't count the ones that  I have done for free but if you count everything total 15 cohorts for last three years before chat  GPT I started this course and it's continuing after chat GPT agents or even anthropics cloud  code if you want to connect I will accept LinkedIn request if you use this URL and send me LinkedIn  request I will accept all the requests that were sent to us now so that's pretty much it let's  get to work. So what we're going to learn is where we are. So we will just do a quick round up  on where we stand today in interview prep or how interview prep is going and what the job market  looks like. Then we will look at what companies are actually looking for the realing really hiring  signals. Then how can you be ready for it? What is the proof of work and what you need to show to  be successful to break into AI interviews? And then maybe I give you some guides, some mistakes  to avoid and then let you answer, ask me questions and me answering. Great. Sounds like a plan.  Let's get to it. Okay. What are the latest job trends? I talk to most of my friends because they  reach out and say, "Hey, we need PMs, Mahes." Uh, and then I say, "I have lot of PMs. That's what  my business is." So then they are like, "Okay, what can you do for us?" So if you take in like  two things right there is this step going in.

3:53 There is this idea that AIPMs are going up. So  a lot of jobs in AI and we have this we showed this data to you last time through trends and my  team will be happy to paste the link where you can find the data. These jobs are growing almost like  at 40 40 to 50% year on year since 2023 and but jobs are also growing but the time to fill this  position is also growing. So it's taking it used to take like four to six weeks to hire somebody  now it is turning out to be a 3 months process just to hire people for the hiring managers and I  will tell you why. The second part of this story is people who are also getting hired a lot are  people in three domains. One they are builders or they know how to code. Those are in demand.  If you understand a domain or you understand infrastructure these are also growing and they  are not increasing the time. You can hire these people faster. So still like four to 6 weeks you  can hire them. The jobs are less but the hiring is happening faster. There's a huge middle that is  eliminated which is generic PMs. These are the PMs who are great at execution, understand all the  frameworks, whatever exponent told them about uh they can put everything under this uh five  post five quarter forces and all but they don't understand or don't have lot of AI experience  in their CV. These PMs are either waiting to get here or building the skills so that they can be  here where the jobs are less but if they have the skills they can be hired. So that's the market I'm  seeing. Uh so just an insight to for us to just lay the land to understand where the world is  and why it takes so much time to understand or get hiring here because one is there is lot of  demand versus supply. The supply is really short and I know all of you will say my should we did  your course right? Uh so there is a lot of supply uh but it takes a lot of time after you do my  course to actually get into this pool and the demand is more but they complain to me this is my  hiring manager's grunt or they they their beef is that we can't produce enough and it's taking lot  of time to fill these positions because they are not getting everything and let's see what are  these hiring manager looking for for 3 months.

6:41 asking you to fill lot of surveys, then asking  you for case studies, then putting you into a loop and then saying that hey we are evaluating other  other candidates also so we'll let you know we'll let you know and then maybe never let you know.  So we looked at this if you had 3 weeks nothing else we'll give you a plan at least to get ready  for this but okay what are these companies looking for highly hiring managers are expecting you to  understand the experience on building on these models or understanding how these model works you  have once one of the you there are three times more likeliness for you to get hired if you have  worked on any kind of LLM and taken to production which is also the top three hard skills that they  are looking for now. So this is like when people were interviewing you earlier they were looking  for cloud skills they were looking for building for cloud first mobile first in 2010 2012 era  this is the new era where they are looking for have you played or have you built things with  the models and if you have it you have a high likeliess of getting hired. Second thing what  they are looking for is they don't want you to talk about AI or describe about about AI. They  want you to demonstrate. They want to see what you have done. Have you done any project? How many  customers use it? What kind of problems you faced?

8:10 And can you actually show them a working thing  with great taste and great attention to detail? Whether it is evaluation, whether it is narrowing  the focus of the problem, whether it is pricing, they want to see real work because that's where we  don't have frameworks for evaluating AI candidates yet. So we will ask you to go and say open  your thing and just walk me through what you did. Okay. Knowing about AI like what's happening  in AI, how cloud code is different than co-work, how co-work is different than open claw. These  are the fluency questions which product managers we think that they should know if they want to  work in our teams because we want them to get the best what's happening in the market today and  then bring it to our company and if you don't have that kind of fluency the chances of you getting  in are really hard. Why? because we don't have people who have this AI fluency already in our  teams. That's why we are opening this position for AIPM role outside and in the interview if you  don't show that fluency then you are not getting in either. Okay. And we expect you to be hands-on.  So now you know what we are looking for and when we are looking for all these talented that hey  you understand fundamentally how machine learning and AI works. We are can you demonstrate like what  you have done? Can you talk about latest greatest trends in a way that impact building product  skills and then can you show us some building skills as well? These are the core things that  I am looking for when I'm hiring PMS and these are all my friends when they tell me like hey I  need a PM right now just make one and send it to me. Some of this sometimes like at least three  out of these is a must-h have skills for them.

10:08 when they are looking for it which I can't fill  right away to be honest with you and I paste these jobs but then on our jobs portal but still  they are open for 2 to 3 months and they keep stay open why what skills what is the AIPM skill set  one is ML literacy if you want to go you want to have literacy so that you can start prototyping  with these tools and understand them without this like Without knowing how to do prompting, without  knowing how these knowledge management works, without knowing how multi-agent systems work,  you won't be able to use the best of these tools which people use to do or build products. If you  understand these concepts, you will be able to use these tools to the full potential. If you can use  these tools to full potential, then you will be able to understand the limitations or this new  agentic world better or build that AI fluency or AI literacy that we talk about. And this this  one keeps moving. But to get here first you need to start here then graduate here by continue  doing playing and seeing the limitations or building stuff and then launching it in hands of  people and seeing or actually understanding what a product looks like. That's the must have. Do the  big mute everyone. Mute everyone so we can move on. Okay then. What's next? Then the question top  of mind for everybody is do I really need to start coding? Is PM becoming a software engineer now?  Like what's the difference? Because you're saying go build with cloud code, run open claw, and if  I need to now code, is that what you're asking us to do? Not yet. We're getting there, but not yet.  We're not asking you to code. The expectation is not to code for PMS but being able to use no code  tools and claude coder or co-workers getting into that territory which is like lawyers PMs are using  every day to go build prototypes. So you should be able to use tools like lovable vo to build your  prototypes. You should be able to build something basic or fix few things in claw code. That's the  next expectation. But not we are not expecting you to take things in production and ship end to end  features which thousands of users can use. That's not the expectation from PMS. So don't spend your  lot of time learning Python or graduating in C++.

12:50 So what is the expectation around proof of work?  So that 3 months window, how can you shorten it by giving me all the signals I'm looking  for there? I'm looking for three things. one have you done certificate are like overrated I  know what I gave you one but at least it's it's I need something which says that you have graduated  or you understand all the ML and AI things which can be a certificate which can be your education  which can be your work experience but give me something that tells me that you understand  things and it's written on your CV so I can ask and verify so if you say I did this AI IPM  certification then I can say okay Did you learn these tools? Do you understand these concepts?  And I can query you. Then build some case studies.

13:38 Write a PD which is a AI first PR which has  I will walk you through a AI PD what it looks like and then I want you to do a prototype which  at least 10 customers can use and then you should be able to do a live demo of a working thing  which customers are using in production today. That's the proof of work. People think proof of  work is just a picture or a PR or my certificate.

14:06 None of that. All of it is your proof of work.  Maybe sometimes I will start with PR. Maybe sometimes I will give you some assignment to  build a prototype or I will ask you to show me one last prototype you did or whatever you build  or I will ask you if you have built a product, can you show it to me or any product that you  ever build, can you actually show it to me? So that's proof of work and in that proof of work  what you need to show is a working prototype, a PRD and a system design write up examples.  This is one proof of work that I can show to the world. This one you can just take a contracts.  So if you have lot of contracts in your company, you can come here, you can click a contract and  you can extract terms from it. And when you do it, you can add new things. We can upload your stuff  and then start analysis. And then it gives you everything that you were looking for. But the  beautiful part is that it shows you how you can edit stuff. That's where continuous learning  is happening. You can click it. It will take you to the page and you can see the justification of  why this was done. What was the citation? where it is taken for and why the answer came the way.  You can even edit the reasoning. That means you are not only training the model with the right  data, you can edit the reasoning. So see I can talk about five things in like 2 minutes to you  because I have built or you have this something that is right right there. Okay. Then what is a  PRD looks like? A PR will look like with a clear definition around why machine learning and AI. a  solution which talks about what are the security and compliance things that you're going to  follow something where you have narrowed the focus so that the team can create an awesome road  map. So you have done prioritization around your stuff and you created road maps that went into  production that team can ship which is what will the MVP will have what's the date looks like what  is the launch what are the iterations after that and then you do you have responsible AI practices  what are you going to transparently disclose what is your data looks like what testing you have  done so all these are good ingredients of a P that you want to show. And the last thing have a  blog ready which is put on some maybe substack but something like this where some validation like  for me Microsoft but you may not have it maybe your last company post a blog like this where  you have gone and shown a system diagram. So if you want to ask me a technical question there is  a system diagram I can talk about it I can walk you through it. So now you can see the system  diagram and system design and any P specific things. So if you have ask me any question I will  take you there and I will walk you through things.

17:09 So now I'm building that confidence that you need  in me that I have not only done projects I have actually shipped stuff. What kind of projects? So  these are good examples of product. So don't show me like a chatbot that you build in this era or  even like a support bot is like becoming a table stakes. So I want to see something like this which  is somebody build an RFP automation agent. What is that? An RFP automation agent is it checks  request for proposal of government companies or government puts RFPs. It automatically checks  that validates whether your company is eligible for this or not. If eligible creates the first  draft of it and then assign different people to review which is an awesome thing. Now this whole  thing the company can now file 10 times more RFPs because you built this agent and somebody built  it and became like got a promotion inside their company. So that's what good example would be. Uh  somebody build a compliance for accessibility. So if you are in UK or if you are in EU there is a  compliance law or regulation which expects all the websites to be compliant to people who can't  hear, see or have other disability. Can you build an agent which checks every website out there or  every agent out there in the new world and see if it is compliant to this regulation? And if not,  then it creates a letter to the company and says, "Hey, I found these problems. I can fix it for  you in $10,000. If you don't pay me, I'm happy to report you because I am completely against this  mission or I'm completely aligned to the mission of building more accessible world. Great. And you  can show end to end and you don't need anybody's permission to build it and you can show results.  And the last is red teaming Robin Hood. uh this is what other team build which was this idea of they  go and check agents. I was talking to somebody at US un University of South California USC LA and  they build a AI trust lab and they are building this red teaming Robin Hood and if you have worked  or if you have done our cohort happy to connect you I'm trying to collaborate with them so you  can be part of that AI trust lab and build these agents. So what I'm proposing to them is building  this red teaming team outside. So any agents that go, we can attack it for security, for safety,  for harmfulness, helpfulness and honesty and build a report card and then show the world that  this is the good agents and bad agents out there and you can do that today by building this red  teaming Robin Hood agent. So these are the good projects to have or the projects that actually  get attention from recruiters or inside your companies. is if you want to become that AI person  inside your company and don't do just chat bots in 2026 I think they are little outdated now chat  bots means you put some documents and you can ask questions on it like the kind of one I was showing  but we have more in that so I will show you next time uh aentic okr is another one it must be live  and likable linkable so you should be all these if you're building it should be live and I can go  and click on it and I can show or experience it if you're going to show it to me if I'm your hiring  manager. Okay. What is getting tested? This is like the main thing I have seen around OpenAI,  Meta, Google Anthropic. I have helped a lot of people like inter prepare for these interviews  in last 6 months and you are going to get one flavor of this which is either you're going to  get they all have some kind of AI product sense AI execution there is a technical round there is  a behavioral round to see if you are a culture fit or not and then there is a presentation they will  ask you to build or a case study they will give you and they will give you some that presentation  thing maybe before or maybe live during the interview. So that's pretty much all the interview  rounds look like and you can prepare. It's going to be very technical. So do not expect or do not  get surprised if they keep asking you technical questions because they will ask you system design.  I've seen people getting system design questions.

21:51 I've seen people ask being asked questions like  hey here is the app that we want to build. Here is a one pager. go build it right now in front  of us. They will and you can get P 0 lovable or whatever you use. So these are the rounds. So what  are they testing in AI product sense? Four things. one, what problem are you are you able to identify  AI problems and how are you going to differentiate yourself today from clawed code, open claw like  agentic systems which are already doing lot of things inside the agent loop for you is your  product building trust with the user or eroding trust. So they want to see when they I'm testing  for production sense I'm seeing is your product building trust or eroding trust with customers  and then obviously evaluations evaluations is not straightforward in AI. So I want to make sure  that you give me those signals in product sense and then how will you make money on it because  there is an ROI problem with AI. We have spent too much money and we have not made a lot of money  out of it. So everything is right now may get out of balance someday. So I want my PMs to at least  have a product sense where they can think how can GTM and start making money for us. So this is what  I'm testing or anybody's testing when they are testing you for AI product sense. So what is your  3WE road map first week get to fundamentals build the literacy build the landscape build what you  need to go and start building basic labs. If you have done our course, you have already done this.  You should know all the fundamentals. So this is our certification course. But if you don't want to  do the course, you can build it yourself and do it with the lab. So you know all the challenges and  limitations of these things also. Then week two, you should start building on your ideas and  shipping things in customer hands. The customers can be your own family members. I want 10 people  to try your thing every day and give you feedback.

23:56 And then you should evaluate it, set your  benchmarks, improve it and iterate and iterate. That should be your week two. And week three, I  want you to start doing mocks. And then the last thing I want to land there is that past is slow.  So this week three onwards I think uh Nancy came to our cohort last time and said this that each  mock interview the successful candidates prepare 6 hours just watching themselves and building  the muscle of not doing many mocks if then one one people look at this metric is how many mocks  I have done and other metrics is how many hours I spend analyzing the mocks and actually filling the  gaps. So when you do the mocks you will realize that you have these gaps in your first side.  You don't understand the fundamentals enough.

24:51 You don't know what is a chunking strategy in  a rag system is you don't understand when the how to evaluate a rag system which is different  than evaluating just your models. So if you don't understand you have to go back and build those  fundamentals. If there are gaps in your product or your project then you have to go and build that  and then keep doing this loop again and again and I promise you it will take 3 months but you will  be there. What are the common mistakes? This is the last thing I promise and then you can ask  me questions. I know I know it takes a lot. Okay then what are the common mistakes? First one I see  lot of people study AI a lot. They get into this mode of just doing courses and courses on Corsera.  They will sign up for every Maven course out there and they will learn lot of things but don't build  anything. Unless you build, you will not know the limitations. You will not have the experience. You  will not know what doesn't work. And people who don't know what doesn't work or what didn't work  for them and how they made it working and still shipped. I need these three parts in your story.  In your AI stories, I want to see three things.

26:06 I did X, it failed on Y, I found Z workarounds  and still I shipped to N number of customers. So this X Y Z and N number of customers is important  for me. And if you have not built, you can talk about the concepts but you will not be able to  land these. And that's the common mistake I see when I ask questions like hey what do you think  that uh we have you used rag I saw rag in your systems what is your experience with rag and then  the person goes on oh rag is retrieval augmented generation it's a process where you take the  embedding models blah blah blah blah blah blah blah but then never tell me like how they used  it where it failed what they did to solve it and how many people are using that And that's what I  want to hear here. And people don't tell me that and that's why they wait 3 months to get a call  back. Another thing I see is a lot of people are working and applying lot of jobs without building  a portfolio. So they don't have those three four things I earlier talked about. They don't have a  live website. They don't have a live project. They don't have a single project where they understand  the complete system design. and they have not solved any customer problems with AI ever. And  if you have not done it, the chances are you know single digit for you to get that role. And  then the last thing I have seen is the behavioral interviews especially in open AI and anthropic  they want to see lot of alignment of your mission to their mission because once they know once  you get inside that company you will be a high demand demanding uh high in demand product manager  and they want you to stick around and one way to ensure that you will stick around is to ensure  that you are moved by the mission especially Enthropic I've seen like four or five candidates  now in entropic where everything is perfect just they are saying you are not a cultural fit because  you have not shown enough that why anthropic they will ask you simple questions like hey why you  want to join anthropic and that question means specially something special anthropic is doing  which meta is not doing which openai is not doing which super intelligence labs are not doing  I think first like this doing idea was very angle good. Second on that doing angle is I will talk  about the constitutional uh training way they do which costs two times more to train a model than a  normal non-constitutional way and then if somebody asks me what is constitutional training then  I can talk about the constitutional principles that Tario has talked about or the paper I read.  So that's number one I will talk about that hey they spend 2x more money with this constitutional  training method and that's why they are one thing second thing I will also land is that this is the  most costefficient company and they have said that they will be the first profitable company so the  kind of focus they give and they are not making an economical disaster by just spending money and  expanding their expense footprint without making any money they have made money they have made  money on the products they shown enough value that people are ready to pay for it. So that's  the other part I would say that security does not mean just securing your models or securing your  training. It also means securing an economic model for the society that we all can find the benefits  and pay for those benefits. So the kind of focus they brought in in AI cost that's the other thing  and the last thing I will say is that we have gone into computer use where we are giving bash  as well as our file system access and they have made that when they build that they build it in  so much trust in the product that whenever it does it it transparently comes and asks you all the  questions it doesn't go beyond it it would have been a disaster like chat at GPT was when it came  out it was just telling people how to do suicides.

30:16 it was telling them how to do chemical bombs but  they release something even more harmful which is has access to your bash and run commands and have  access to your file system but still able to not erode that trust and that's why I would say if  if there is a future where machines are going to take on most of the work that as humans  we're going to do then I want to join that company which is doing it in the right way and  paying the cost and the last thing I will also and is the transparency angle. They are advocating  to government to put auditable transparency. So anytime there is a fault or anytime there is  a gap or known failure, they have to report it. So currently no lab report failures in their  models. If they fail it, they silently keep it to themselves. And entropic is the first lab which  is trying to go and say hey let's build a law which allows everybody to transparently inform  that if we failed then why we failed and how many failures have we got because of these four  reasons I think anthropic is the right company and I personally believe that building trust comes  by building secure AI and without trust we cannot sell anything in AI and that's why I want to join  anthropic so cover those four or five things, land those all ideas and now the interviewer can  ask me anything. How you think about pricing? How about the cost? They can go into my ML and AI  concepts. They can talk about the projects. So I landed all those four five points that I was  looking for when I was I want to interview or I was giving you advices on how to interview.  Okay, your next step. Nobody told me to switch the camera. minus 10 to all the people at least  in the cohort. Now I realize that I didn't switch or I didn't stop sharing when I was on the board  all the time. Okay, then one decision right now what will you be building next week? So I think  this is new era in 2026. We want more PMs who can build and if you don't have anything that you're  building next week I want you to start building and for that I can help you. We are building this  stuff which is allows you to build with cloud code. Gives you everything that you need to know  and install open claw. We tried installing open yesterday for 2 hours and only like 20% people  can actually install it. We're going to show up again and help people install and get ready  with these tools. So that's what you need. And then we need to understand this new agentic loop  that we are getting into. Okay, we are starting a course tomorrow which is a good four weeks course  now where there will be six labs which will help you build with these latest greatest tools like  cloud code and open claw. Uh we will have we have 11 mentors as you heard that awesome mentor which  talked about the passion she had about security.

33:19 She's our security experts in house. Dominic and  then we have 10 more. And we will have three guest speakers who will talk about how to automate your  job search who will talk about what it takes to break into AI product science interviews. Nancy  and then I'm looking to get one of my VPs to talk. I'm just getting a confirmation so that they can  tell you the real other side of the interview process which is why are they waiting, what they  are looking for and what it takes to break into the inside their companies. Okay, if you do that  then we have two courses. One is the first one which is starts tomorrow. So you can enroll on  Maven. I think it should be top two or top three courses on Maven today and the next one starts in  May and you can get both of them for the price of one. I think that's the coolest deal our sales put  in. I need to read this because I didn't know what the deal was. So in summary, there is this idea  that either you have an awesome building skills or builder skills and or you have this idea that  you have built lot of agentic stuff and you can land lot of agentic ideas then you can succeed in  this K-shaped market where either the high domain low number of jobs but if you have that domain  expertise you will be hired with some AI flavoring or you build the new agentic capability. But it is  a long process but sign up for that. Then I talk about what companies are actually looking for,  what a proof of work actually means and then we talk about what is being tested and how to get  ready for it along with common mistakes and I gave you some next steps. Hopefully I will see you  tomorrow. I am starting this new thing where we will talk about interviews where we'll talk about  labs and what the new world of agents look like with the agentic loop and then build these agents  also with claude code and all. With that said, I want to take questions on scrum first. So maybe  Arja can you give me a link on scrumbler? Let's take questions on scrum and then maybe few hands  if time permits. There are more resources if you continue to follow us here. You can subscribe for  all our future sessions. And here are our free sessions recordings, our Substack, our YouTube  channel. And one thing that we love the most, the next role. This is one of our team uh previous  cohort member Charu built this. It automates your complete process, reviews your CV, gets your CV  ready for any job. So just put upload your CV, upload the job description and it formats  the CV. And if you click the next button, it can automatically apply for you  also. And with that said, thanks a lot.