Video Of The Week: Marc Andreessen at Stanford Business School

I watched this interview recently and really liked it. It’s long, around 50mins, but it’s good.

#VC & Technology

Comments (Archived):

  1. Vineeth Kariappa


  2. John Revay

    I think I watched it a few months ago, I think I was lead to it from a post on…..The one comment that I remember the most from the video…was that during pitch meetings….he liked it when routinely his GPs challenged the CEO with questions…and he/her sticks to their guns…and come back w/ something like….”I have been thinking about this for several years…and this is why” vs rolling over saying that it was a good comment, or they needed to think about it/ consider it….

    1. LE

      I think that’s a combination of two concepts one is “people who know their shit” are attractive as well as “people who don’t take any shit” are even more attractive. [1]That said there can also be a downside which is reading to much into someone who simply talks a good game.vs rolling over saying that it was a good comment, or they needed to think about it/ consider it….Look if you really know about something you should be able to run circles around someone who hasn’t really thought much about it. That goes without saying. But otoh the circles you are running might not even be right because the person listening probably doesn’t know anywhere near enough to even challenge what you are saying at all. Who am I to challenge Arnold if he tells me about wine or JLM when he tells me the best place to get [some specialty food] in Austin Tx?That said when I talk to people about something that I don’t know about obviously one of the signals that is rated very highly in my brain is how quickly and confidently they answer a question. What that means is “I’ve run into that before and I know exactly what I did that worked before”. My dad used to see a handyman or tradesman and he would say “I see how quickly his hands move so I know he knows what he is doing”.[1] (Example: Stories about Gates liking it when people kick back at him..)

  3. Richard

    Fred, how big a role do economies of scale (number of partners and support teams) play a role in venture capital?

  4. Emily Merkle

    <no attention=”” span=””> :)will ask Marc for synopsis hah

    1. JamesHRH

      You don’t need to watch it to know its crammed full of smart, applied wisdom.

      1. Emily Merkle

        <known> πŸ™‚ we chat on various.

        1. Carl Rahn Griffith

          Name-dropper πŸ˜‰

          1. Emily Merkle

            dude. did I say a Name? Fred started it…:)

  5. William Mougayar

    I totally agree with Marc that “the product doesn’t sell itself”. Some entrepreneurs confuse the “product attracting users itself” vs. “making the product sell something” that has dollars attached to it. And often that is not the same product. This is especially true in the case of large networks of engaged users with a consumer bent on it. The product users use and the product users buy are not the same. And these users aren’t the same either.You can think of the double-decker Bus analogy to understand this. (I just came up with it)First the passengers start to fill-up the upper section of the bus first because it has the view, it’s open, and it’s free. They are the users. The bus is driving around and everybody is happy. Then, the driver runs out of gas, has no money, but can’t ask these users to pay-up because they were told the ride was free. These users are very intelligent, they are talking about all kinds of things, but they refuse to pay a single cent because they were promised a free ride.So, the driver changes strategy, puts a new sign on the bus that says you have to pay now based on some value inside, and he opens the lower deck for those paying customers. The lower deck is a totally different experience. The passengers there don’t care about the view, but they care about everything that’s going on with the passengers above. They care about what these upper-deck passengers are saying, what they want, and what they need. They start to gain intelligence from these users above. Then, some of these lower-deck users start thinking of new products and services that they can offer to the passengers on the upper-deck, e.g. drinks (it’s hot), and food (they are hungry now), and they even sell them hats and other accessories. So, the bus driver unlocks the internal stairs that connect the lower part to the upper part, and he lets the lower deck users go upstairs and sell stuff to the passengers above, as long as they give him a % of the transactions.The bus stops finally to gas-up. The whole town has heard about that bus, and people are lining-up to get either on the top or lower decks, depending on their needs. The bus driver starts buying more buses, and going to new towns, etc…

    1. LE

      Nice analogy (coming from self a self appointed analogicalgist).So, the bus driver unlocks the internal stairs that connect the lower part to the upper part, and he lets the lower deck users go upstairs and sell stuff to the passengers above, as long as they give him a % of the transactions.That’s an interesting idea actually. You engage the brain and resources of your existing paid customer base to come up with ideas to make money off the “top of the bus” riders.

      1. William Mougayar

        Cool.I would add that the driver needs to approve these products/services, to ensure they are in line with business model & objectives.

    2. JLM

      .Love the analogy. Very well played.From now on you shall be known as “Double Decker William” or just “DD”.JLM.

      1. JamesHRH

        Double Bill?

    3. awaldstein

      Hmm…feels like a club more than an open community model.

      1. William Mougayar

        Not intended to be closed. Isn’t that how Twitter, Facebook etc operate? They sell advertising on top of our discussions.

        1. awaldstein

          You are talking about a curated, club model where approval is the toll booth and the business model.Twitter and Facebook approve nothing. Anyone can play. Anyone can advertise (within legal bounds).They are media models for simply that reason. Both are transactionless by nature. Both are great models though.

          1. William Mougayar

            They approve and create Ad units & sell them to brands.

          2. awaldstein

            You or I can’t place ads or promote content on FB and Twitter–rhetorical question of course.

          3. William Mougayar

            Yes you can! For a few bucks you can promote your tweet, and you can boost a Facebook share so it’s seen by more people.Ironically, one could see it this way: twitter / Facebook / even google contribute to creating the noise, then they charge you to rise above it. Nice model πŸ˜‰

          4. awaldstein


          5. Emily Merkle

            actually, yes you can. πŸ™‚

    4. diymanik

      Thanks for the analogy. This is the exact model I’m trying to employ with my startup’s product and your analogy really clarified the strategy for me.

      1. William Mougayar


        1. Emily Merkle

          sometimes a startup offers not a “product” per se. Service + value.

  6. sigmaalgebra

    So, Marc and his firm have a shortage of managers like Sheryl. Well, Sheryl is in the role of a COO. Hmm. The US is just awash in people who have done well in COO roles in K-12 education, as college deans and presidents, in local, state, and federal government, in various parts of business, and in the US military managing hundreds, thousands, or tens of thousands of people. There’s a ‘shortage’?So, Andreessen Horowitz has people looking for technical and/or engineering talent. Hmm.So, let’s see: The last story I heard was that their talent people insisted on good knowledge of and lots of experience with the programming language Python on Linux for serious production computing. Hmm. Right, can either upchuck now or read a little more and upchuck then:A little exploration shows that Python, e.g., at…was the invention of Guido van Rossum and, for the most popular version, CPython, is written in the programming language C, is quite portable to various operating systems, and is interpretive and dynamic.So, based on portable C, it is limited by what portable C provides and, thus, has to be single threaded (in a world where can get single Intel processors that run 8 or 16 threads) and has to struggle with the given and primitive C memory management. Since it is interpretive it is likely ballpark 10 times slower than a compiled language. Since it is dynamic, it makes static analysis of the code, e.g., desk checking or proof reading, difficult to impossible. The language is heavily recommended for scripting or even used as an interactive calculator.One of the main advantages of the language is a collection of additional software to help in getting data from the Web, e.g., via parsing HTML, essentially ‘screen scraping’.So, no, Python is not for serious production computing.Learning Python? It’s basically just a slightly different flavor of syntactic sugar borrowed from much of the history of programming languages, especially Java, Microsoft’s C# and Visual Basic .NET, back to Ada, PL/I, Rexx, Pascal, Fortran, Algol, etc. A good feature of Python is that it avoids the deliberately idiosyncratic and too sparse syntax of C and C++ (originally just a preprocessor to C).This common stuff of hiring nontechnical HR people to select software experts, computer scientists, and the most promising contributors to information technology in the future is a total bummer.Sorry A16Z talent team. Have to be reminded of the story “If Carpenters Were Hired Like Programmers” at…with more at…Marc, when you want the ceiling of the chapel painted, you don’t leave the selection of the artist to people who know about painting only from watching some house painters.Marc, you might want to upgrade the technical qualifications of your talent team before they filter out the people you most want.”We like to say that our day job is crushing entrepreneurs’ hope and dreams.”. Well, Marc, it’s not at all clear that your firm is very good at picking the next Microsoft, Cisco, Yahoo, Google, Twitter, or Facebook. More generally from the ROI figures in…apparently on average US information technology venture capital is not good at picking winners.”It’s actually, after you have been in it for a while, the thing that’s actually fairly easy to tell is, will this team and company be fundable by a top VC?”Of course it’s easy: The sound in the Sand Hill Road echo chamber is simple and easy to learn: Large potential market and traction significant and growing rapidly. Team eager to sign the standard terms.But venture capital is all about the really exceptional, and the Silicon Valley echo chamber, the average sound on Sand Hill Road, has no chance of seeing the exceptional until too late.So, look at 3000 deals a year, narrow down to 200 to take seriously, and invest in 20? Gee, those 20 should really be something, right? Nearly all of those should be hitting grand slam home runs, right? They’re not? Gee. Are you sure you are digging in the right place? The oil patch learned to use not just seat of the pants intuition but the fast Fourier transform; is venture capital willing to look at technology?”If there is one thing that’s frustrating in this job that every VC deals with, you know, you miss most of the big winners, right?””It’s like the thing all the top venture firms have in common is that they did not invest in most of the great successful technology companies, which is an incredibly frustrating thing.”So, right, “the top venture firms” really are poor at picking winners.Right, and thus an entrepreneur should not be surprised when a VC firm is not interested; neither should a serious entrepreneur be discouraged by opinions of “the top venture firms”.”There’s kind of the surface level stuff you look for, you look for huge market, you look for, you know, differentiating technology, and you look for incredible people.”You look for “differentiating technology”? Really? Do you, can you, evaluate technology? A Ph.D. committee, the editorial staff of a leading peer reviewed journal of original research, the NSF, Darpa, ONR, NIH, etc. can. Can “the top venture firms”? Who in a top venture firm has published peer reviewed original research in technology, is qualified for a tenure track position at a good research university, could be hired as a problem sponsor at NIH, Darpa, ONR, or NIH, or could serve on the editorial staff of a good journal of original research?Subra Suresh, from MIT, NSF, and now at CMU, can evaluate technology in mechanical engineering. Who on Sand Hill Road can do anything comparable in information technology? Count them all on one hand?For the “incredible people” just how to define or recognize those? That they were writing C code when they were 12 years old? At a good university, that won’t even go very far for a BS in computer science. In an auto analogy, writing C code at 12 is like replacing an auto radiator but nothing like being a mechanical engineer. In a construction analogy, like building a dog house but nothing like making fine furniture or being a civil engineer or an architect.”Genius and courage”? There is the line from ‘King Solomon’s Mines’, “Your courage is based on ignorance and does you no credit.”. For genius, just how to recognize that? Okay, for information technology, what is a good model? How about John von Neumann? If so, then what was it that von Neumann did that indicated genius? How about his proof of the Radon-Nikodym theorem? What did you notice that was brilliant in that proof? What about Hilbert space? What is the genius there? What about game theory? What games does that theory apply to? How does that theory work? What is the role of random variables in that theory? What does the theory have to do with linear programming? Who on Sand Hill Road knows the answers to such questions?In information technology, what about C. Shannon. One intuitive view is, the number of little balls that can be placed in a big ball. Okay, what is the role of those big and little balls in his theory?Information technology is about ‘information’, right? Outline what we mean by ‘information’ and how we can use some information we do have to estimate or approximate some information we don’t have. What role do you see here for Hilbert space?Information we do/do not have? How about accurate ad targeting? How to attack that? With just intuitive heuristics or with something more solid and promising?As your partner Ben learned at Loudcloud and HP, there is Open View and Mercury Interactive that are important for the crucial work of real time system monitoring and that can report data in essentially real time on many variables at data rates from a point each few seconds up to many points per second.Okay, using this data for system monitoring is necessarily and inescapably either just some intuitive heuristics or a statistical hypothesis test, and we should prefer the second. For that work, we’d like to have tests that are both distribution-free and multi-dimensional. Where can we get some of those? Would that be “differentiated technology”? Someone who cooked up the first and only version of that would be a “genius”? If so, why; if not, why not? Ergodic theory going back to G. D. Birkhoff, E. Hopf, H. PoincarΓ©, A. Garcia, etc. has to do with measure preserving. What is that? Could it have a role in distribution-free, multi-dimensional hypothesis tests? Can see also Dunford and Schwartz, ‘Linear Operators’. Yes, Jack Schwartz was long at Courant and working in computing. Was he a “genius”?”Failure sucks. Success is wonderful.”. But in Silicon Valley, in academics and research, success sucks and failure is wonderful! So, after Google, a theme was to invest in Ph.D. program dropouts. Looking for “incredible people”? Hmm.Look for people who “never quit” and have to see that “in their background”? How about if they quit their education?Good that you got out of the deep, sticky, smelly stuff at Loudcloud. Not good that you got into it.That’s enough.Last night the code with the bug I fixed compiled. Time to run it!

    1. Rick

      I come from programming low-level languages and I must say the higher level stuff, like python/django, is all about productivity. There is no need to worry about where the bits are and how they toggle when you’re wanting to focus on the application domain and object models that will implement it.

      1. sigmaalgebra

        the higher level stuff, like python/django, is all about productivity. Of course.I’ve programmed in assembler for three quite different processors.On small projects assembler can be several times faster than compiled code. Once my assembler code was 6 times faster than the corresponding Fortran code. But such a speed difference is commonly due mostly just to clever register exploitation, and, really, a good compiler should be able to do still better than a programmer will bother to do.Also, there is an old story about using assembler: Once a project gets to much in size, say, more than 10 pages of assembler code, a compiled language such as Algol, Fortran, Pascal, Ada, PL/I (can see a little of the syntax in SQL and Rexx!), C, C++, C#, and Visual Basic .NET can be significantly faster than assembler. Why? Because too soon the assembler programmer needs various cases of ‘support’ for, say, memory management, I/O, exceptional condition handling, usually too difficult for an assembler programmer to develop for themselves. Old story.E.g., I have to believe that code in assembler for a Web site would be slower than one on Windows Server, SQL Server, IIS, ASP.NET, and Visual Basic .NET, assuming the sun didn’t burn out by the time the poor, struggling assembler programmer got the first version running.Maybe that was the problem with the first ObamaCare Web sites — the things were written in assembler? :=)!Still, there remain needs for assembler, C, etc.Now speed in the simple sense of assembler execution is not the main issue, even in keeping down hardware costs; instead ‘architectural’ issues dominate.And, for the applications software, right, ‘productivity’ is the big issue, and assembler is like trying to dig a new Panama Canal with a teaspoon.But I was objecting to the statement of the recruiter that they wanted people with expertise in using Python for large scale, production computing. The short answer is that they are looking for something in the empty set — Python should not be used for such computing.Instead, use C#, Visual Basic .NET, or some such. For a good compiled language on Linux for commercial, Internet applications, I don’t know of one. I won’t consider C++ as a solution if only to the basic fact that, due to some obscure aspects of the design of C++, a significantly large C++ is nearly guaranteed to have memory leaks through version 10.0 or so. So, the programmers and the whole project is constantly swatting dumb memory leak problems instead of doing the real work.For ‘managed code’ on .NET, due to the quite advanced work in memory management, mostly get just to f’get about memory leaks.Memory leaks? From all I can see, at the Web site of Business Insider, some of their JavaScript code commonly has memory leaks and/or infinite loops — too many of their Web pages pin my CPU busy at 100% with the virtual memory file growing to infinity. If people can’t get memory leaks and infinite loops out of JavaScript code, then we should try to avoid such development tools.At one time an interpretive language had some advantages in programmer productivity, but my view is that now the good compiled languages have caught up. There’s no longer much trade off; instead we can have the best of both.

        1. Rick

          I’m not sure what your argument is against using python in a production environment? By production you mean delivered to the user… Right?.I can see by your comments you’re very concerned about speed. But in the business world you always tackle speed issues when they arise during use. For example… You don’t optimize a routine until you’re sure the user will actually use it often. If you optimized every feature agreed to for making someone happy in a meeting. The cost of optimization would drive project costs through the roof. Only to find that person replaced with someone who has a new set of pet features.

          1. sigmaalgebra

            I’m not sure what your argument is against using python in a production environment? By production you mean delivered to the user… Right? Well, first, the recruiter meant high volume, high performance production. There, broadly a factor of 10 in performance for Python, a ballpark estimate based on Python, in particular the most popular version CPython, being interpretive, compared with a compiled language is a bit much to give up. Or, if paying for the capex for 10 times as many stacks of servers in columns and rows and the floor space and the corresponding opex for the electric power, then that factor of 10 starts to cost real green money. I can see by your comments you’re very concerned about speed. In a production environment where performance is important, a factor of 10 is speed it a lot to give up. Anyone should be concerned.Sure, likely we are broadly in agreement. The remain issues are a bit subtle but still important. But in the business world you always tackle speed issues when they arise during use. Here “always” is a bit strong, and “during use” will often be a bit too late. And we don’t want another case of a hobgoblin of little minds, e.g., spending big bucks now to avoid a chance of a problem, more likely criticism, down the road. Or we don’t want to go broke buying insurance against possible problems in the future. Right: We don’t want ‘premature’ optimization; we also don’t want a long walk on a short pier. For example… You don’t optimize a routine until you’re sure the user will actually use it often. Broadly, correct. But usually we should mostly be able to tell what will get a lot of use, assuming our application does at all, during design and not have to wait for code ‘profiling’ during use.But, sure, sometimes we wait: E.g., in my application, I’m keeping the data in SQL Server and in early production will usually get that data directly from SQL Server; in high performance production, my design and architecture has other approaches that can ‘scale’ to serve the world, if I’m that lucky, with options for little or no direct, live use of SQL Server; with my architecture, these options just replace some small chunks of current code within the architecture. And, right, as you mention, what options I pick when will depend on where the performance bottlenecks are as my site sends 0.1, 1, 10, 100, …, tens of thousands of Web pages a second, if such things happen. But, these options still have nothing to do with, or to recommend, Python at any level of scaling, over what I’m currently using.And we do not want to start across the Atlantic Ocean in a birch bark canoe and then try to build a seaworthy vessel while we are in a storm half way to Europe. For my project, I’m not writing prototype code; I see no advantage in doing so; instead, I’m writing code for at least early production as the first code.Here’s my main view: I had to choose between Linux and Windows. While I’ve had a lot of experience with several of the most important operating systems in computing, and while the main points of Unix and Linux are no doubt very similar to what I’ve had long experience with, actually I’d never used either Unix or Linux. Instead my then recent background had been with IBM’s MVS, IBM’s OS/2, and Microsoft’s Windows 2000 and Windows XP.I believed that Microsoft has a lot of highly motivated people working on documentation, fixes, and improvements in Windows, the .NET Framework, the .NET languages, IIS, ASP.NET, Windows Server, and SQL Server. And if I need support I can get a lot for free on StackOverflow, sometimes, given a good question, also for free, directly from expert Microsoft people on Microsoft fora (once I got some really good help on details of writing polymorphic code, and once directly from a high up guy in SQL Server land), and later, when I’ve got bucks enough to hire, etc., can get paid support, hopefully with an ‘account executive’ who will make the support people make me happy. In total, that situation seemed to me to promise to save me time and effort, although in the long run, not necessarily money, compared with Unix/Linux. Sure, if eventually I get to have billions a year in revenue, then I will set up a Linux team of maybe 100 people, and they will know every line of code in each instance of Linux we are using, keep up on system installation, system management, system monitoring, bug fixes, device drivers, performance, etc. Mostly until then, Microsoft stands to save me time, effort, and, compared with a 100 person team, money.Or, with Microsoft, then (and now) I could go with the operating system I knew instead of adding yet another largely redundant wheel to my operating system truck.Then in Windows land, quickly I learned from a friend and also lots of documentation about Microsoft’s ideas for ‘managed code’, the common language runtime (CLR), the .NET Framework, and the .NET languages. I saw that for a language I could pick either C# or Visual Basic .NET and that they were, for nearly everything, and for everything I needed to do, equivalent ways to get to the CLR and the .NET Framework and otherwise differing only in the flavor of syntactic sugar, and there I didn’t like the C-like idiosyncratic syntax of C# so selected Visual Basic .NET for it’s more traditional syntax, instead. I’ve written enough C code and put up with the K&R fun with deliberate obscurity, but I don’t like it.So, now for my project, essentially all my production code is in Visual Basic .NET. Learn it once and use it many times in my project.For CPython, I see no advantage, and a lot of major disadvantages, in using it in the main production code in my project.Yes, there actually is a ‘managed code’ compiled Python, IronPython; that version of Python would overcome a lot of the disadvantages of CPython.Why? In simple terms, for being easy to learn, write, read, and work with, it’s tough to beat Visual Basic .NET. For my main code, Python just doesn’t have a better hammer, wrench, screwdriver, or nail. For my main code, nothing, not any version of Python or anything else, promises to be better than Visual Basic .NET.Now, could I make good use of some version of Python? Possibly: There are a lot of software packages for Python, e.g., CPython, and maybe some of those packages would be quite helpful for some peripheral parts of my project. E.g., I might like Python for grabbing some HTML and parsing it. Okay, maybe in a few weeks I’ll do some of that, with a version of Python. Fine. ‘Learn’ Python? What’s to learn besides a slightly different flavor of syntactic sugar beyond what I’ve long since learned in, say, at least 10 other programming languages? Nothing really new to ‘learn’. For that work, the disadvantages of Python, even CPython, would not be of concern. Some of your observations apply here: The ‘performance’ of Python would not matter.CPython could be nice for a lot of little specialized tasks, but it has no advantages for, stands to have no role in, the main code of my project. Keeping that code all in Visual Basic .NET is tough to beat.Scripting? I do that now with Open Object Rexx and get a lot of utility from it. On Windows, I should convert to Power Shell for its good integration with Windows, and likely will when I get a few days. Scripting with Python on Windows? Likely not except maybe for IronPython.To be more clear, Visual Basic .NET is easy enough to use for writing code that using it for the core code of my project is not at all a case of ‘premature’ optimization. Or, for Visual Basic and the .NET Framework, for the core work of my project, I can’t think of anything in computing now that could be easier or more ‘productive’.Broadly, CPython, just in being interpretive, and I would add, being dynamic, is not good for high performance, production code. So, the recruiter wanted knowledge and experience with Python for high performance, production code — he wanted someone with experience using a teaspoon to dig a new Panama Canal. It would be no good for me or anyone else to get involved in such an uninformed, misinformed, hopeless situation. No thanks.But, I confess, maybe on Linux there is nothing competitive with either C#, Visual Basic .NET, or even IronPython, so that on Linux CPython might be the least bad alternative.Really, .NET was a LOT of work, a lot of which is good; I can believe that for the intended uses of .NET, Linux will have a tough time competing.I’m sorry to drag this out; I thought that all this was totally, grade school Computing 101 baby talk crystal clear; we’re not talking P = NP here, guys; here I’m doing all I can to be still more clear.

          2. Rick

            While I enjoyed your posts. I do believe from seeing how you provide an entry into a decision tree but never provide your conclusion. And that you constantly put speed over every other issue. I think you are suffering from OCD (obsessive compulsive development). You are going to do what you want the way you want even if it undermines the project..I’ve seen this many times in the past. Engineers get so spun up in their own mind that the world around them fades into oblivion. I’ve seen them sleep on the floor at work just to work extra hours to prove that their way would have made the application a few cycles faster. Even though the project was canceled weeks before due to cost overruns..I do admire your passion but I would say that you should consider putting forth some effort to keep your life in balance. BTW I know this because I was in your boat before. Now that I’ve walked away from production coding I see how wasteful all that time and effort can be.

          3. sigmaalgebra

            While I enjoyed your posts. I do believe from seeing how you provide an entry into a decision tree but never provide your conclusion. So, it’s completely clear: You have in mind what you anticipate I am thinking, doing, meant, and wrote but, no matter how often or clearly I wrote, just will not pay good attention to what I wrote and, instead, stick with your prior belief that is just not nearly correct.For the end of “the decision tree”, I will repeat and summarize, thoroughly, in nine points:(1) Windows.I had to pick exactly one of Windows or Linux and picked Windows. Quite broadly that is not a bad decision and for my background was a good decision. All that is crystal clear from what I wrote.(2) Managed Code.In recent years, Microsoft has worked hard on their ‘managed code’, CLR, and .NET Framework as the foundation for applications software. So, on Windows, that is also the foundation of my software. All of that is totally crystal clear to anyone with a good background in computing at any time for the past 5+ years.(3) Visual Basic .NET.On that Microsoft platform, there is no programming language significantly more powerful or productive than Visual Basic .NET. The main alternative is just C#, and it is different only in its flavor of syntactic sugar, maybe a little better on a few niche features, maybe reflection or lambdas, which I don’t need or want to work with anyway, and maybe not even those. So, for now, for using the Microsoft platform, Visual Basic .NET is essentially the best decision possible. This point is totally clear to anyone with even a superficial understanding of software development on Microsoft for the past 5+ years.(4) More on Visual Basic .NET.At least for now, Visual Basic .NET has a good collection of the best ideas in programming languages back to at least Algol 60 in 1960, that is, draws from Java, C++, PL/I, Ada, Pascal, Fortran, Algol, etc. It’s tough, for practice, now, to expect a language to have a better collection of features. This point is totally clear to anyone with a good background in the history of programming languages.(5) Programmer Productivity.Python and especially some of its software packages, can be more productive for some niche applications. Something similar can be said about Matlab, R, Mathematica, KnowledgeTool, and more.Still, for selecting just one programming language for commercial programming now, Visual Basic .NET is equal to the best or nearly so for everything but niche problems.So, for my project, I selected that language for everywhere it is suitable which is for all the more important software in my project. There is no significantly better single language. There is no part of that work where another language could be significantly more productive. So, net, my selection and use of Visual Basic .NET is fine.In particular, Visual Basic is so good that an interpretive language would not be more productive.(6) Execution Efficiency.Visual Basic .NET is essentially ‘compiled’ and, thus, is ballpark 10 times faster than an interpretive language. So, an interpretive language has a high cost in execution efficiency but, for the main work, brings no offsetting advantages.For large scale, production computing, a factor of 10 in execution efficiency nearly always has to be a far too high a price to pay.(7) Python.I might use CPython for a few niche problems, but from the above it is totally clear that CPython is not suitable for my main work.IronPython could be considered but for my main work does not promise to be better.So, for my main work, CPython is not suitable.More generally, my view is that for nearly all production software on Windows now, there is nothing significantly better Visual Basic .NET and CPython is not suitable.(8) The Recruiting.The recruiter wanted expertise in Python for large scale, production computing. Maybe on Linux, without C# or Visual Basic .NET, although there may be some efforts on Linux like there are efforts to port the .NET Framework to Linux, CPython is about the best choice available.If CPython is the best choice for such computing on Linux, then that is a very sad situation and, thus, one I would not want to get involved with.But, for my main programming language on Windows, I should have no interest in Python, and I do not.(9) OCD.That is a psychiatric diagnosis and for me is uninformed, misinformed, incompetent, insulting, and wildly and totally wrong.Do I really ‘love’ Visual Basic .NET? No, not really: I frequently fume at the low level of the technical writing on Microsoft’s MSDN including for Visual Basic .NET. E.g., the writers have yet to understand how to describe a ‘correspondence’, i.e., a ‘mapping’. Indeed, my view is that the low level of that technical writing is one of the main bottlenecks throttling the future of Microsoft and the rest of computing, literally. If the writers can’t describe the products, then programmers can’t effectively and efficiently use the products, and usage of those products will have a tough time growing.Am I in love with Windows? Not at all.With Microsoft? Nope.Instead, just as I made totally explicit above, I am using Windows, the CLR, the .NET Framework, and Visual Basic .NET because they are the best decisions for me for my work now. And I believe that for now these choices are the best decisions or nearly so for nearly all of applications software.In the future I hope for a much better operating system (e.g., with good solutions to the continuing, grim security problems with some serious security architecture and guarantees) and system installation, management, and administration, a better programming language and development environment, and much better technical writing.You have not actually been reading and considering what I wrote and, instead, are in some old ‘programming language religious wars’ and ending with insults instead of thoughts. I will not respond to you further.

          4. Rick

            I wasn’t trying to insult you. I was just relating something that I’ve seen many times over the past 20 years..Again I wasn’t trying to insult you. I have no reason to do that! Also I don’t fight any language wars. I’ve moved on from being a programmer to things I enjoy more..I wish you the best and hope the system you’re working on turns out great and makes you gazillions of dollars.

  7. KelvinMPadilla

    Who am I to challenge Arnold if he tells me about wine or JLM when he tells me the best place to get [some specialty food] in Austin Tx?

  8. johndodds

    Very good and almost totally lacking in jargon. I think the two are connected.

  9. Vasudev Ram

    I had seen that video a while ago. Don’t remember how it was, but I do remember that this one (also about Marc, and Dave McClure) is good:Marc Andreessen & Dave McClure: The Future of Startups, VC’s, and Technology…

  10. Carl Rahn Griffith

    He’s the only person to ever block me on Twitter. I’ve been there a tad longer than him.Strange cove πŸ™‚

    1. Emily Merkle


  11. Rob Underwood

    His description of the valley in late 90’s – the focus on selling, deals, and “go big or go home” about 8 minutes in – really resonated w/ me. I was at Pandesic (a story in of itself) at the time, and I recall vividly the focus on deals throughout the valley and “ecosystem” — and as Marc implies it definitely felt like the sales folks held more weight than the technologists, and that seems very different today.

  12. jason wright

    Uber/ Lyft technologies + driverless cars = rapid decline in personal car ownership?

  13. jason wright

    did he answer the ebay/ paypal question, or not?

  14. Niv Dror

    For those of you that doesn’t have 50min watch, make time. Otherwise, I wrote an in depth summary if Marc’s talk at Stanford on the @datafoxco blog:

  15. pointsnfigures

    agree on marketing. if i were a kid today and wasn’t technical, I’d go to business school and really learn statistics/marketing/operations. huge part of the value chain that straight tech founders discount but are an essential part of the business. Cheapest capital to get is revenue.

    1. Emily Merkle

      stats / ops in soc.sci as well. studied PhD .edu psych & I/O psych. has served me well.

  16. Jude

    His essays are better than Paul Graham’s in my honest opinion. Shame he took them down, god knows why.

  17. Ovidiu Schiopu

    Smart. Articulate. Thoughtful. Insightful. Whether you agree or not it adds to the conversation.

  18. MFishbein

    It was interesting that Marc criticized the “x for y”/”Hollywood approach” yet sites that 35 of the top 100 games in the ios store are clones. So if it’s working, why is that bad?