Thriving in Engineering

Michi Goetz is the Director of Technical Program Management at Celonis, and he writes TPM Breakdowns, a Substack on the realities of program management.

In this conversation, we discuss:
1: Why status reports turn red two days before launch, and why calling it red early is a gift, not a failure
2: What his survey of 250 TPMs revealed about where AI adoption actually stands
3: Why bad underlying data breaks AI's usefulness before it even starts
4: His "TPM boardroom" method for pressure-testing decisions with AI personas before a meeting
5: The three stages he sees TPMs moving through, ending in what he calls an "AI architect"
6: His weekly voice-note practice for building a personal decision record
Michi’s details:
TPM Breakdowns Newsletter: https://michigoetz.substack.com/
LinkedIn: https://www.linkedin.com/in/michi-goetz/

What is Thriving in Engineering?

Software engineering isn't just strategy and sprint plans. It's ambiguity, pressure, trade-offs, and people.

Thriving In Engineering helps you navigate all of it.

In the newsletter, Alex Ponomarev distills 25 years of engineering leadership into practical frameworks to help you lead with confidence, stay calm under pressure, and build thriving teams.

The podcast goes deeper, talking to engineers, leaders, and builders across the industry about what AI transformation actually looks like in practice, past the headlines and hype. One guest at a time.

Hello and welcome to Thriving in Engineering podcast. In each episode I talk to people

across the software engineering industry about AI. Today we have Michi Goetz. Michi, welcome to the

podcast. Please tell our listeners about yourself a little bit. Yeah, thanks Alex. Thanks for having

me. So my name is Michi. I'm leading an technical program management team. I will start with a fun

fact about me. Originally I'm aerospace engineer. I tried my own luck in funding companies and I'm

with a company called Celonis like who was in the process intelligence space since 8 years as

an early employee and I've served there as product manager lead and CTO office and ended up leading

a TPM team and driving 30 plus programs for the product and engineering organization which spans

from the US to Spain, Germany and India. That's that's pretty cool. And uh as far as I know,

Celonis is a massive company that you grew with and you've seen through this hypers scale growth,

right? So you have tremendous experience that I'm sure will be very interesting to our leaders.

I really love your newsletter TPM Breakdowns on Substack and one of your first post like you you

have a lot of memes there and one of the first post with this meme with the green status that

is not actually true. Can you tell a bit more like the story behind the meme and yeah what's the joke

there? Yeah, let me start like why why I started to write on Substack I ended up speaking before

incorporating the TPM role to our organization to a lot of like TPM leads specifically in the

US. So it's not very well known or penetrated in the European market. And I have a lot of like

inquiries like from mentoring from other TPMs. This is why I started the Substack. And one thing

why I believe tech is so great as well is like because we have the greatest memes and sometimes

in our daily life we have to we have a lot of challenges And a good meme could also break the

ice. And about the green the status report. This is a I think a a classic thing that everyone has

seen in across like product engineering engine leadership like you get this green dashboard

and then 2 days before due date or launch date it falls into into red. So the joke itself is about

the program status. It's more like a social aspect not a technical aspect. Because everything is

green because nobody like wants to be the person to turn it into red. Although like the work is on

fire and I think like what I mentioned like before like every engineering leader has lived that

moment and you're in a review it's green everybody is parting ways having a coffee but you already

have this suspicion when you go into details that you know that separate conversation it

feels like red or whatever you use like amber it's maybe not to be rogue on it and I think it lands

because like no matter like which role and tech you had it could also be go to market role like

everyone has lived that experience and I thought it's a a great start to write about it and a great

like memory to have it back because everyone can literally relate to it and a green status

or a red status or an amber status is something which is daily bread and butter for TPMS and this

is like what the Substack is also is also about and it also has I think inherits a lot of things

about how people are communicating with each other because sometimes I'm a strong believer is a good

thing to call it red because it's better to have a conversation like early on rather than later and

everyone is going into a code red or cold yellow however you you feel it and it drives also clarity

and maybe what others call like hindsight it was good that it's being called red rather than green

throughout the data. Yeah, that actually makes a lot of sense. I can surely relate. Can you tell

tell us a bit more about your journey at Celonis in general and how how your role changed in the

past few years as we all are going through this AI transformation. However cliche this phrase

might sound. Yeah. So the one thing where like where it started was I I officially

joined as data scientist solution engineer. And this was only like for a few weeks until

one of the founders like reached out to me because I was like developing some some benchmarking tool

and so forth and we had a deeper conversation and then I I joined the the product team. You

can imagine it. It was a bunch of 30 product and engineering members who who built a software and

the software is all about like how can you improve the processes. The interesting point was at that

point in time so imagine it being 2018. So you had this big lag of Europe not moving like to

the cloud or moving like SaaS products like where let's say for Europe quite early in the like US

was very fundamentally upfront the goal was like to move our like on-premise application to um and

cloud SaaS product and we did that in under a year like including migrations and the hiccups that you

have at the beginning and you need to make a lot of choices. So my part there was to focus on what

we call like analytical components which is now moved the studio. So you can imagine that you have

a process explorer where you can see like how your processes are really running. And then like also

embedding like new products on the UI perspective talking about AI. So in the product like we always

had AI we just call it in that point in time like a machine learning workbench. So you can imagine

it like being in Jupyter notebook where you can like execute Python. Which I think like if you're

realistic like when you look into nowadays like still it is in the background that many

have like Python helper scripts in your Claude Code or Cursor or whatever you use to to fuel

agents and so forth. And I then switched my role to what we call like engineering operations

which was at the end of the day in CTO office. And I did that for two and a half years. And

personally I was missing one thing. This was what the tech is all about or the software development

is about working with technology and when you run like some sort of CTO office or possible

like CEO office like you're very much like into this process vein and like what can be improved

and so forth but I was like personally like missing tech and we had like reached a certain

saturation point but we still have been growing and the good thing was I had time to reshape what

we really look into it and as a product manager and also like when being a CT office I was like

every other month in the US is I spoke to a lot of like TPM leaders directors like TPM IC's and I had

this role like I was following it longer also from literature and people you get to know on LinkedIn

and I spoke to them to to make a case in order like how does the role function other companies

and how does it fit into our organization because I know the organization by heart and it helped the

organization to set up that role and then build it out from scratch. So I had like two people at the

beginning and now it's like more than 20 people. And we're focusing heavily on cross-functional

programs and the differentiation is we don't spend so TPMs don't spend time in backlogs or

in sprint meetings or in those kind of things. is like more like certain cross-functional programs

that could be migration programs could also be like a new product line where many teams are

involved and so forth and orchestrating that at the end. The question like on AI I think

it's very like let's say nuance because the mission at the end stays the same so you still

have a technical complexity people need to make decision like be product be it engineering I

mean we can have a discussion like who is getting compressed or from a timing perspective. I think

the opportunity that you saw in the last like one and a half years is that AI helps you to overcome

certain things that have been time consuming you know like tracking, summarizing things,

where people have spent so much time in it and if you think of the agent world and agent is just a

it's rule based it's like summarize me that and given that I think like people should start to

like embrace the opportunity that they have. And there's one I think like one key point that that

many should not forget is like the underlying data no matter if you use like Jira, Aha, GitHub or

Linear or whatever you use this needs to be clean because if an agent is calling something or wants

to get something then this needs to be clean and then this amplifies no matter like if you're in

product, if you're in engineering, if you're TPM, if you're QA or like in that role or design. If

you have a good data foundation, then you're set up for for success. This helps a lot like to to

amplify AI in that that sense. And I was heavily for AI like we what we did at the end of the day

is a little bit the journey. So we started to give the team the freedom to experiment. And then

we did an hackathon like to break it's not about the use cases that you develop for a hackathon.

changing the mindset that more people starting to adopt it. And this then also like helped to

drive the team into a certain strategic direction to to embrace it and use AI in that sense. But it

it's a different way of working where I think like everyone at the moment tries to find out

where it fits best on everything that you see externally and how can you adopt it to um to

your company. I think we're still in this not in a consolidation phase but like more in a still in

a little bit an early experimentation phase. Yeah I think all of us are in all organizations across

the world with things changes changing so fast and to switch from macro to micro how the work of

TPM changed .how your work changed now with all the tools do you get to work with the technical

side now with tools like Claude Code because we engineering managers, we're lucky to now be able

to work on some issues, merge some pull requests, and get our hands dirty. Yeah. So I recently wrote

on Substack as well and we also did like an survey on like 250 TPMs with other like peers TPM peers

from different companies and what is like very what was very interesting to see and I fully can

relate to that is that many are still in this beginner phase. So they're not intermediate or

advanced. So everyone I think has solved the mystery around summarization. How do you do

meetings and how do you like the simple parts I think the the hard part comes also as well and

what was also an outcome of the survey is that your company needs to also give you access to

the tools. What we what we have done at my company at the moment is like we use Cursor

and I think at the end of the day no matter if you use Cursor or if you use Claude Code the

the baseline is always the same so you can solve like this like simple things as you call it like

very very much what it on what I mentioned before on the other side it also exposes like how your

data needs to be very good so if you don't have a risk registry if you don't have proper Jira setup

and this gets fully exposed like like it makes your systems like very exposable to it because

you need to to feed it and then AI cannot help you fundamentally what is changing I think is that

you can take the time and no matter like if as mentioned like if you're in product or engineering

no matter if you if you're like what you're doing with AI, like this busy work gives you time back

and this I think is an opportunity for all roles like in a tech company to focus back on what is

important like this strategic judgment, like reading the room. AI cannot help you in what

are what is your you can it can come up with a stakeholder map but still humans need to talk to

each other to make a judgment and also have final calls. Yeah. AI also changes the cost of showing

up unprepared. So it means like when you come to a meeting and you prepare it like 30 seconds before,

you there's like there is no excuse anymore to come not prepared to a meeting. So it's not a

choice anymore if you come prepared. I think it's getting like more in mandate that and if

you have a meeting that it becomes prepared. Another chance is well to do an audit. What I

also like like wrote about like in December about like you can also audit your meetings quite well

which which meetings do I literally know because the information breadth is like there like way

heavy and this is what I like most because it gives a different aspect on it. Have you seen

situations where people prepare for the meeting by reading AI prep? I've seen situations where

people are just channeling what AI told them and they're essentially a talking head and the

judgment is missing. It's a very interesting phenomenon I think over the past years. Yeah,

I think there's a tactic also where I could not help. I think like it's like how you prepare your

or like prepare your meetings, let's call it that way. So I'm a big fan of pre-reads and I think

what AI has helped us all is like to move away from this like 20 pages PRD engineering design

doc or program memo and so forth. It helps a lot in order to prepare way quicker like a one/two

pager where people can comment before the meeting. So as one asset so that you don't need to answer

the same questions in the meeting and so people can like that's why I love like I'm exposed here

like that's why I love Google Docs for instance like you people can comment you can prepare that

beforehand and it helps you like to not rush into the day before when no one is reading it pre-read

anymore but to prepare it like 5 days before so the meeting gets also like way more efficient and

in the meeting like to make that also sufficient what I like also like you can let no matter like

if it's a design doc and PRD like you can it you can let it rev you from others and what I mean

by others is not like potentially a human you can also let it judge by a by a reviewer let's call it

that way so I have a use case which is called a TPM boardroom and there's a skeptic and CPO

and CTO and if you do like a program revenue AI council and then the first round it's arguing for

What is the agent actually? It's itself like [laughter] like a like a itself and then they

haggling it with with each other like who is right and then can form that into is it good bad or is

it like the skeptic will still be skeptic or give doesn't give a go and this then also gives

you a very good angle which questions do you want to have answered before everyone is leaving the

meeting and I think there's two ingredients like a pre-read plus you know definition of done or

acceptance criteria areas for that meeting to come to the point where makes it more easier and the

the human interactions in meetings like way more efficient. I love this approach and when I was

talking about quote unquote simpler things, that's what I meant because you can have a very simple

meeting summarization and that's it. But what what you're talking about it's a totally different

game. It has to be engineered. It's a process. And who who knows better about the processes

than TPMs, right? I like my personal opinion is that that's what us managers in general will be

doing more. So less of the mechanics like hey here are the action points after the meeting sending to

everyone asking for updates and this and that coordination more about actually engineering

the processes. But that's like that's my point of view. I'm curious how do you see the role

of a TPM going forward with all the tools that we have now and the tools that are coming like

is this role going away or is it changing? Like what do you think? Yeah. So I think the role

is not going away and is not being replaced. I think the role in itself like will fundamentally

change and I think like there are three like three stages that everyone is going through like one is

like more this you're a tracker of something then you orchestrate something and then you become

what I call like an AI architect at that end I think similar things will happen like also and

to the TPM what I mentioned at the beginning like you move away from this tracker perspective of

moving things because I think this is like what is so-called table stacks at the end of the day what

TPMS have always own is this cross-functionality and also the processes on how this things work and

this is like what you can like very systematically also codify. So when you look for instance into

something like a launch review then you have a launch review readiness and there you have

also rules like your test coverage need to be like this. The communication needs to be ready,

the releases needs to be ready, the release notes needs to be ready. These are rules that you can

very specifically give to an agent and say is my test coverage above a certain degree? Yes. No. If

not then you can intervene a human. What I think as an main point why the TPM role is not going

away is when you think like what TPMS sometimes do they wear different hats. So they translate

this like technical complexity or into business requirements and vice versa. And you wear then on

the one side sometimes also a product head or an engineering hat and all of that because you have

this cross-functional context. You might now argue that the cross functional context is given because

like the organizations are building a knowledge graph or a contextual graph of the organization

but at the end the TPM work includes like three main things like one is like also influence like

without authority and this is like also comes with a stakeholder management, also identifying

where things are really a risk. So you can have a risk which is on paper fundamentally a risk but

with someone like who you need to judge that risk is it like really that big or is it like really

really that heavy but still there's some human judgment in the room because as long as humans

are involved in the software development life cycle and you still would need a TPM because

there is this cross-functional area there because like no one could cover it but I think the role

will more shift like to more strategic program management. So where you have this big programs

across the organization when you roll out certain things you know like it's you see that a bottom

up let's say like agentification and I but I see more that the strategic program management will be

more coming to the cornerstone where you need to put like real thinking into it what does it mean,

what is the impact, and the good thing is you can also document all of that knowledge so in now in

before times you have documented like all your sprint tickets and I think like TPMs will also

start like what is the decision record and this makes the agents also better and you can also then

spend more time into strategic judgment. Yeah, I can't agree more to be honest. I don't think

anyone is getting replaced unless like your job is purely repetition. But that's that's

an issue in itself. I also think that we're this is this is this paradox that we're having. More

things are getting built, more programs are being rolled out. Somebody has to manage them. Somebody

has to be accountable and AI is not going to be accountable after all and yeah judgment calls

will always be on humans as well. I think one niche here as well is like who is managing agents.

So the thing is like there's a huge opportunity and I think there's also a discussion going on

in in the product like industry when you read on Substack or Medium is like who who will manage the

agent and I think this is agents this will also be an opportunity like also for like each of the

roles like to to doing that because we all have been there like coming back to the green status

like who is it about to blame like you know do you then start give your agent a bad performance

which actually might be a nice thing because it's like the agents going in the performance

review end of year and getting their bonus or more tokens. I don't know but you know there's a lot

of like opportunity and open questions. Yeah, I think all of us will be managing agents. It's like

it's inevitable. I can see this happening in in many organizations already. Especially with Cursor

and Claude every team from I don't know from TPM to DevOps like throughout the organization what do

you think will change for leaders in general because we talked about TPM specifically but

like leadership changes as well not just because of the tools but because of how we

work is you you said that yourself. The nature of work is changing. So this inevitably leads to

uh the nature of leadership changing as well. Do you see uh the the transition happening there?

Yeah. So I think like one interesting fact is where to start is like what is a leader? I think

like AI unlocks it the capability that everyone can become a leader because currently like when

you see in some engineering teams like you have this and what I also an advise like listen to

the silent person in the room like they they also getting empowered to become leaders and I

think the biggest productivity gains from AI for leaders are not in the execution I think they're

in the work that happens before the execution starts this is like problems structuring,

stakeholder alignment, risk identification, also scenario planning because agents are not there

like to forecast. They do a job and get you an input or an output no matter where you sit. And

I think like leaders who use AI to to think harder before committing to a direction will consistently

outperform those who use it to go faster into one direction before that has already been chosen. And

I think this is like exactly a little bit what is I think happening to everyone is like this

upstream strategic work gets way more important. And this is I think like an big opportunity.

Everyone should look into it it no matter like if you're an engineering leader or if you're an

IC wants to become a leader. And I think this line blurs like very heavily and you see that

in the last year like should I do an IC career path or should I do an M career path. I think

this line like gets blasted away because you see the tendency that many go back to an IC role and I

think there will be a huge shift. But I still like see the evidence of managers but both have this

leader perspective in place and maybe as a short second point I think there is what is also worth

noting is that is institutional institutionalizing judgment. So I think you have to have one concrete

pattern in the decision record. So that's why I think like architecture decision records or

organizational decision records become like more important because they identify a design rational

and postmortem learnings and they solidify the organizational law knowledge and might also feed

into like let's say general LLMs that help other organizations to become better in order to not

fall into the trap or even this myth around like could a one person build a 1 billion company I

think if someone cracks it that this patterns that so in happening in the software industry can be

solidified into a general LLM could help in order to do that. But before do that like I think an

organization who doubles down on decision records, design rationals and everything is documented

systematic. I think this will be such a fuel for agents how they act in the company context. And I

think this might be a new kind of organizational asset that might help leaders who start to

document their things or have this documentation already in place or have an GitHub for decision

records. This will like empower them fundamentally to make the right decision so that you have a lot

of leverage as a leader going forward. And I think this might be an advice like also for every leader

to start documenting your stuff. You can also use a voice on what you have done every day.

Yeah, I think it's great advice to summarize what you just said in simple words. I just had

it spinning in my head. Everything you can do anything now with AI tools, but not everything

at once. So deciding what to do and building a framework for these decisions going forward,

that's the real differentiator for leaders. Yeah, I think what you mean is like output is nearly

free. But building this coherence on documenting the things that happen in the organization if

everyone is doing it then you build a system and standards that compound over time. Exactly.

Yeah. Yeah. Any other advice to the leaders and engineers and TPMs listening to us? You already

mentioned documenting everything as much as possible throughout the organization. Voice notes

at the minimum maybe anything else. I think what will definitely not go away the next 12 months

is a promo cycle. No matter like if you have it once a year or twice a year and everyone is in

gold rushing the last four weeks. What I have done the last 12 months. The challenge is like with AI

like it's like a train that is very super fast at the moment. What I can recommend is like to record

voice notes every week. So make a calendar invite in your thing and just or make it a meet even or

record it on your iPhone or which phone you ever use every week and then let it put into a markdown

file or a document and let it transcript. Just by making your own decision record for yourself and

documenting it properly helps you in your career moving forward. I think very heavily no matter

like if you're in an engineering leader or like also an IC but it gives you clarity so many don't

have had the time to document what they have done and now it's like the entry barrier is so low by

just recording it sounds very weird though when you do it the first like two or three times but

try it out and then let me or like Alex you like know on how that works but this is like I think

my top advice at the moment. I do that like since 5 months now and it helps a lot like in order to

save it automatically and then give you a Monday briefing or even a Sunday briefing matter like

how your week starts but it it helps a lot like in order to to circle back on what you could do.

I love this advice so much. I'm because I'm doing exactly the same thing. I started writing a diary

just like a personal one but inevitably work and personal stuff is tightly related at least

for me. So I end up putting everything there and what I started doing maybe like last year

there's an app called Granola that can take meeting notes and so on but they have a like

an iOS app maybe Android as well so you can use it to take voice notes. And I'm what I'm doing

is and it works offline and so on. So, and there are other apps. It doesn't matter which you use,

but like I'm doing this at the end of the day when I'm driving from the office home,

I do reflection. It costs no time because I'm driving. It's just instead of listening music,

I have this and then exactly as you said, I can go back, I have all the summaries, all the

transcripts. I can go back and see the progression of what I was thinking, what I was doing.

I think it's such an amazing advice for work and also for personal life. It's just wonderful. Thank

you. Yeah, 100%. Don't be scared to do it. Yeah, it definitely is weird initially, but then you

get used to it and you don't feel like a person talking to to oneself. Michi, thank you so much

for joining today. I think you have tremendous experience and I highly recommend all of our

listeners and viewers to go and check out the Substack, the TPM Breakdowns Substack. And yeah,

thank you for joining. Thanks, Alex, for having me and yeah, thanks for having me. Thank you.