Discoursebeta

Steven Sinofsky AMA

Steven Sinofsky

I am an advisor, board member, investor, who writes a lot. I started off as a computer science / chemistry double major back when almost no one used a computer in chemistry and the computer science department was more like a math department. I did undergraduate research in the very earliest IDEs (called “syntax directed editors”) which made me want to be a professor. So I went to graduate school and promptly dropped out when I realized that programming was more fun that writing about programming. I interviewed at a “software company in the Pacific Northwest” when no one in my family had ever been to Seattle or had any idea what a software company might do. Everyone I interviewed with seemed to know everything about everything so I took the job and then spent almost 24 years there from 1989-2013. I worked on the first C++ compiler and tools—that research. I did a weird job as “Technical Assistant to the Chairman” and became something of an internet-is-a-big-deal clarion call. I then spent over 10 years working on Office from the last 16-bit version to the biggest redesign of the products, introducing Clippy, email, web versions of the tools, and web-server collaboration along the way. I then moved to Windows where we shipped Windows 7 and Windows 8. I’ve managed teams from single digits to five digits in size, managed teams in a dozen countries, and hired more people than I can count from college to join Microsoft. Along the way I lived in China to combat piracy. I taught MBA students in Boston to gain a different perspective. And wrote more blog posts than I can count. I have a lot of stories. I love to compare and contrast old and new. I love thinking about the stories of product development and how they were told and retold over years. You can read many on "Hardcore Software" hardcoresoftware.substack.com—a tale of my time at Microsoft serialized on Substack and a my long form posts on current events in tech. I’ve advised, invested, or served on the board of quite a few companies and worked closely with a16z on several companies where I served as a board partner representing the firm’s investment. This is an AMA about these topics. Thank you!

Discussion

Dhruv Gupta

Hey Steven! It sounds like when you were coming out of undergrad taking a job in tech was a risk/not high status. For much of the mid to late 2010s (and again now) it feels like tech is that cool, hip thing that all the best people coming out of the top universities want to do again. How different was it before the status? Do you ever wish the industry could go back to that (or do you think, in some ways, the people pushing the frontier have never cared about the status it brings)?

Abhay
In reply to a comment

+1 to this question. Leaving tech seems like the wrong type of low status thing to do though. Do you have any perspectives on how to seek relatively low status but high value things to do in tech today in 2026?

Trace Cohen

You’ve seen a bunch of major tech shifts firsthand at Microsoft. With everything happening in AI right now, what actually feels different this time, and what feels like the same movie playing out again? If you were starting a company today, what lessons from those past shifts would you keep in mind?

Erik Torenberg

You were at Microsoft when the internet was happening, and responded accordingly. What would you tell large software companies today on how to adapt to AI?

David Booth

do you think large organizations of five digits are obsolete now with AI? Some economic theories suggest that Coase theorem should predict a biomodal-ification of the economy: giant intelligent producers on the one hand, and a long tail of intelligence consumers on the other

Zach Davidson

thanks for doing this steven! know you’ve managed teams ranging from a few people to tens of thousands. what’s one management practice that worked at small scale but failed as the organization grew?

Elena

If you were in charge of a university computer science department today, how would you run it? What kinds of classes would you mandate / what's 100% obsolete now?

Steven Sinofsky
In reply to a comment

There's always a risk of being the general fighting the last war and with my passion for history of tech I really need to keep that in check. By far the biggest difference and one that matters the most with AI is the difference between being completely deterministic / algorithmic and being predominantly stochastic. This too isn't totally new. The history of computer science is simulation. My college program required a number of simulation projects--meaning use of RANDOM() and understanding that each run of a program would be different but understanding why. We have long lived with very important "models" such as financial and weather. With those people understood they were best case "predictions" but no one advertised them as certainty. AI is being used for a lot of things (like search or answers) where the expectation is to be algorithmically correct. Some of this has to do with the way AI chat has communicated with humans -- the presumption of being an oracle is based into the style and language the GPTs use which I have a problem with in general (see the book "The Media Equation" and also CLIPPY!). Then there's the way the AI companies talk about their own products in the third person and use anthropomorphic language. There's a long history going back to the 1950s in AI using such language and of course Unix did as well (child, parent, sibling, kill, spawn, etc.) But in the context of 1B people using a product it lands differently. The mechanics of product development and market mechanisms, well those seem a lot a like. Yes money is required now and the whole software revolution was relatively light on spend. I suspect this will change over time. We're on a decidedly google trajectory of data center-centric view of solving problems. That can change just as the PC changed the idea of a mainframe.

Brent Liang

Hi Steven! Curious about your perspective on how frontier AI policies should evolve regarding open-source models (opportunities, pitfalls, etc).

Steven Sinofsky
In reply to a comment

This is difficult because any time you go against the current conventional wisdom you either sound old or just wrong. There is a reality that you can get started with vastly less by way of resources today than you could as a saas company which was less than a pc software company which was less than a pc hardware company which was less than a mini computer which was less than a mainframe company (see the trend). I think it might take ever fewer people to start a company and show impact. But quickly there is some mean reversion with respect to selling, support, communicating, and doing anything at scale beyond the initial creation. It is difficult to imagine this changing by two orders of magnitude for example.

Steven Sinofsky
In reply to a comment

Build a a whole new product where you don't assume the interface is about formatting a document. The vast majority of surface area in Office is dedicated to "debugging" the format of a document. AI can do this very well (it should be trained on a corpus reflecting the goal though and I think demand and enterprise-specific data will matter.) When the web came along we started the web versions of the apps and the service side (SharePoint) as whole new things. The innovator's dilemma part of this was both internal and external. I always felt the company resisted the web versions and saw them as "low end" which was not what I might have hoped. I knew it would be a long haul and might not work but sitting here today it is clear that installing gigabytes on a PC is not an asset that it once was.

Abhay

If I were figuring out what to do today, I would try to figure out whats low status (like you joining Microsoft back then was) but also potentially extremely valuable. Do you have any sense for what some of those areas may be today?

Steven Sinofsky
In reply to a comment

This is a great question I love. In my own education and then hiring students one thing I saw was a constant tension with what to teach versus what was happening in industry and where the job market was going. I was at an "elite" school which saw this tension both internally and externally. Internally the tension was over what the faculty (many of the first generation of any CS professors and founders of the field) felt (and created) the goal of computer science was versus the new people who wanted to be relevant. My first day of CS100 we were told "we will teach you how to program. To do that we will use a language we created called PL/CS which is like another language you never heard of called PL/1. We won't teach you BASIC or Pascal. That's because we teach you to 'program INTO a language, not IN a language'." This was always frustrating. C was just happening and I wanted to learn C. Taught myself but it always seemed weird. In contrast in Chemistry they were teaching us 100 year old organic and quant, and not the latest in molecular chemistry or whatever was in some new article about curing cancer. Later CS classes were about the fundamentals across the field: machine architecture, OS, DB, PL, Algorithmic complexity, numerical methods/simulation, and so on and a lot of theory that still seems questionable but the origins of the department were math. We were some of the first CS students though not to have to take electrical engineering. MIT students still did. I saw that difference later and had fun conversations with Chuck Thacker (co-inventor of ethernet, Xerox Alto) who built stuff AND wrote code. Today I think there will be a rush in some departments to stop teaching coding. I liken this to the rush to teach Visual Basic or other HLLs in the 1980s versus ASM and C. I think it is clear that most every programming will not "hand code" but to be the best at understanding the way machines work really knowing this foundation seems important to me. Another way to think of this is that majoring in CS should equip you to build the next tools not just use the current tools.

Steven Sinofsky
In reply to a comment

I think that since the 1980s open source is part of the fabric of computer science. It isn't going anywhere. Honest and a bit spicy I think the biggest "hack" to open source has been how the licenses that arose allowed for a company to take an existing open source project and put it in a data center and use it to build a whole company and never release any changes. I wish there was something that could be done about this because I feel it misses the spirit of open source. This was why Microsoft was so against open source though -- the viral spreading of work was also anti using it. You could use open source but then everything you did HAD to be open source. So we should really find a way to solve this because the incentive is to release something, get traction from a community, then own/control it via a data center interface to the public project. Google of course pioneered this. I don't see any pitfalls in terms of national security or capitalism to maintain open source as a competitive part of the market. Only upside.

Steven Sinofsky
In reply to a comment

This one is easy :-) At the top level of the company pick a flagship process that touches a lot of people and maters—not the most mission critical one but important—and rework it entirely with AI. With the OG PC many companies did budgets as an example or compensation. With the internet at Microsoft many teams created collaboration sites internally but across the company we engineered the processes for external support and even how we sent out product information to the 10000 person field sales org using web technology. The thing not to do -- set up a committee to define rules to use AI before anyone is using AI. There are some very important things to educate people about like uploading information particularly customer information that should not be done at all. And of course "human in the loop" on any and all generated materials. This isn't just for fiduciary content (contracts, financials) but even for the "random" testimonial or white paper.

Abinash
In reply to a comment

I think yes. Because writing code can be easily done by LLMs, but correctness, performance, and robustness can't be determined without the help of an experienced engineer. To get there, you still need to know the fundamentals and institutions.

Chris Clark

With the shift of AI agents pulling data from different APIs and executing tasks behind the scenes, what do you think standalone 'applications’ or 'app stores' will actually look like to consumers in five years?

Steven Sinofsky
In reply to a comment

Yes! Not all children just like all children should appreciate music but not every child should take violin lessons. I'm not a parent but I think parents take different approaches as to what to expose, encourage, or "force" and that's a parenting choice.

Steven Sinofsky
In reply to a comment

Piracy was definitely a cultural and legal challenge much more than technical. We didn't treat it as a technical problem and worked all the angles we could but ultimately it was actually a strategy problem. China was not going to allow software IP so long as they didn't have credible products in market. So we were competing with a strategy couched as culture and legal. By taking any action it was going to be used against us. I don't know what we could have done more or less of when you're managing through sovereignty like that. It is not unlike what we find today with the EU and the DMA or other regulatory stuff going on there.

Steven Sinofsky
In reply to a comment

Good one. It isn't clear to me that we will have a world where there is so much pulling of stuff from everywhere. I tend to think this is how computer people view building solutions but not how people trying to solve the whole problem see things. I am not a believer that all apps devolve into headless databases because the database was never really the hard part of most problems. The hard part is building the workflow and user experience. SAP is a database but it is also 30 years of encoded knowledge of how resource planning works. Many people who see things as connectors really just want to extract data for their use. No doubt about that. And many who just want to use a headless database just find the experience not what they want. This is very common in enterprise fields like reporting and visualization where doing custom work is the norm. It is why every data-reporting/financial products ends up exporting to Excel for analysis. But still to get all the information in there and to have it work as a system requires the whole experience. I'm over simplifying but I think we see in real time the challenges and risks associated with everything connecting to everything as the first order solution.

Steven Sinofsky
In reply to a comment

The biggest difference I experienced is really understanding that when you’re a team of 5 you are really a team of one with 10 hands and you are working as one and sharing with each other as though you have a neural connection. There’s really no filter. Even if you had one after the first time your work sees customers or users then any filter goes away—that’s the essence of team building which is having that winning experience together. This breaks down at some level but definitely a 100 (that’s also the most number of people you can really know in any size organization of thousands.) By then even the people you know well are not sharing with you in the way that even that same person shared with you at 5-10 people. It is a huge bummer. You really have to work at seeing through the filter and not faulting people for using a filter. You’re stuck in a management bubble no matter what. I worked with people I knew from the first day at the company together and sometimes you managed them and sometimes they ended up managing you. Our friendships today all transcending the work. But we clearly had moments of friction due to scaling. It was especially noticeable when for example I was the manager of a person I’d known for 15+ years and then they were also going through the same challenge with managing people they knew for 10 years or maybe even hired from college. This is a reality of scale and was always difficult emotionally for me.

alaska

as software becomes more abundant, modular and malleable, do you think we lose something valuable? is there something to be said for a globally homogenized experience of, say, an operating system, vs. a future where ai is tailoring everything to you specifically

Tunca Üçer

Memos were Microsoft's operating system. Now that AI can draft the memo, what is the one thing that still has to be written by the person in charge?

Vineet

There’s lot & lot happening in AI space. Would it make sense to step back a bit, get strong on basics (writing code, mvps) and dwell into AI wave or tinkering with AI, building mvps, learning practically is a good option? And how to overcome the overwhelming buzz - LLMs, RAG, Inference, Evals, Harness - this is becoming like a never ending loop, we start with something today & tomorrow something new comes up. Would be glad to hear your thoughts sir.

Steven Sinofsky
In reply to a comment

It is interesting to take each of those and try to extrapolate them individually: Abundant - I think this is great and amazing. We’ve been incredibly constrained in our collective ability to create more software. Most companies in software maintained their position less through innovation than through the ability for a competitor to even build something in the same category. So more abundance does mean more innovation and less “just because they won before they keep winning.” Now that’s not always good in terms of innovation, but it will be different. Modular - Software modularity has been the St Elmo’s Fire of the field from the earliest days of Fortran. Everything was about creating more modular and interconnected parts. At some point I think we just have to recognize that modularity might mean many good things but it is also an unnecessary constraint. In the physical world where there is a high cost to having single suppliers or parts that break and require an exact match thus fixing the levels of abstraction in the machine to what they were when designed, software can rethink abstractions and build new layers. These layers allow new ways of solving the problem and approaching solutions which is a net positive. Modularity is related to the first point in the sense that software was so difficult that the only hope was to reuse something that someone else wrote and debugged. But in practice that seemed like an frustrating constraint on the future. Malleable - I wonder if software is more malleable or not? I’m not sure. It can be. But does that always imply something good? It has been said that real programmers spend 90% of their time on 10% of the problem and I always thought that was because software was “too soft” and people did not hesitate to redo what really didn’t need to be changed. It was too malleable maybe? I think on net, every improvement in tools makes for more software and so far that has shown us how we need even more software than that!

Steven Sinofsky
In reply to a comment

Hot take. I think if you’re literally asking AI to draft a complete memo from a bullet point or sentence then do not do that and definitely do not distribute it. It will be very low value. Conversely, if you do write something and then say to AI “make it better” you can very easily get in a doom loop where you just make it too long, handle too many objections, and lose track of what you wanted to say. The sweet spot is drafting your memo and decided for yourself where it is weak — do you have specific objections, fact errors, overlapping strategy points, etc. and use AI to improve those but like a review with a person on the team do not ask for a rewrite but for thematic feedback. What I worry more about is that people will take anything longer than a phone screen and get a summary. No one ever accused me of not writing or not writing enough and it took me a while to realize even as some “big boss” people just didn’t read whole memos. They looked at the structure and wanted to read what was relevant to them. That bugged me a lot but was also a reminder that most people want to talk big strategy but by and large think locally. Related I never did “executive summaries.” I was an exec and certainly saw how many F500 companies and government had a routine process to do executive summaries. I really didn’t like those because once it is there almost no one read the rest. AI can turn everyone into a “my time is so valuable I only read the summary” people and much is lost. My view is that writing is at least 75% for the author to drive clarity in their thinking. The more senior or cross-functional you are the more critical the writing process is. If you skip it you’ll feel fast and agile at the start but will hit a wall down the road when all the stuff you didn’t think through comes back to bite you. It is also critical in a big company for teams and alignment, not to mention new people joining teams, to know some big picture. I never mastered getting people to read everything though.

Steven Sinofsky
In reply to a comment

There's no magic answer to this except you do need to get stuff done and be prepared to redo it later. The space is not going to be settled for a while and will still be changing after that. Think of all the HTML frameworks that we've gone through in 20 years or how the server side is gone from CGI -> VM -> Cloud -> Lambda -> etc etc.

Pareen

You ran developer programs and platform ecosystems at Microsoft. What actually made developers commit to a platform? Was there a particular group that was sticky?

Marcos Ortiz

Thanks for doing this, Steven. Here is my question: What could be the hardest problem you would love to see being solved for a new company out there?

Sign in to join the discussion.