I want to start a daily (or as close to it) thread where I can reflect / discuss things I have realized as life changes. As a recent grad, I have certainly noticed the shift into industry to come with some growing pains. I am also fortunate that I work at a startup where every day has growing pains. I’m hoping to use this thread as a chance to not only get some advice on how to grow myself, but also to think through some of the challenging issues I’m dealing with at work. Feel free to follow along!
Welcome @jsnapoli1 - look forward to reading your posts!
1 Like
(One Reason) Why Product Development is Hard
I have been thinking a lot about the why and how behind making products. Interestingly, so much of my masters program this week has touched on product development, and I have received some good knowledge at every turn, seemingly. But with all of this information around product development, why do people still struggle with managing and setting up a roadmap?
A lot of it is procedure. We tend to like the “fast start” option, where we take something that “worked over here” and apply it to this new, completely different project. In reality, tailoring your processes to the project on hand may slow you down initially but speeds up development substantially in the future. We have found this at Zonit by running software tests on every push, and leveraging Claude to be able to reference these tests and respond to them accordingly. That is just one example of how having a process tailored to the project (a custom fixture for testing our software) can save substantial amounts of time.
Another thing I have noticed companies struggle to balance is innovative products. Customers are good at knowing what they want. But, especially in the data center space, I feel like it’s hard to convince someone they want this brand-new, revolutionary product that they have never seen before. This tends to lead towards incremental improvements on technologies we have already seen. This isn’t always a bad thing, but it does make a company begin to think about what happens when they run out of improvements. Furthermore, I have found it stifles the creative thought of engineers, as we don’t try new things.
If anyone has any insights, I would love to discuss. I am new to the world of product development and product management, and I feel like the scope can become overwhelming. Any tips are appreciated!
This is a tough balance to find - reusing/leveraging what we’ve already done vs. doing new stuff.
I think part of the solution is to be agile, where you own the integration points and it is possible to incrementally and with relatively little pain pull in new technology/tools/workflows as it makes sense.
If you don’t own the integration points, then it is painful to swap in/out large chunks as you don’t have enough control.
1 Like
Agreed @cbrake . I have also struggled to find solutions for non-technical folks. We discussed yesterday putting Product Requirement Documents (PRDs) in Git so we have version control. Ultimately, we went with storing them in our Nextcloud instance because Git was “too technical” for sales to use. I understand it, but it is frustrating, especially as a “highly technical user” (am I allowed to consider myself one of those yet?)
The Value of Transparency
In reality, how much information is absolutely top secret in a company? Like, what would the downside be if all company communications (message boards, PMs, emails) were 100% transparent, where everyone had visibility to what was being discussed? In fact, I would argue that all of the unproductive discussions that occur within company walls (gossip, complaints without intention for resolution, and the like) would entirely disappear if this was the case.
I am not proposing that companies enact a surveillance state on their employees, I am talking about something even more radical - What if everyone in the company shared everything they did on a day to day basis, no matter how big or small they thought it was?
What if, inherent to the company culture, individuals shared every interaction they had? Certainly company knowledge would circulate much faster. Certainly people could hop in and fill in missing holes in conversations people may have had. Imagine how much more productive a day’s work would be if you didn’t have to guess who did or didn’t know about something.
This came up in a discussion today: Should companies disclose everything. “Hey all, I made this part number”, or “Hey all, I had this discussion with this client today”. It would certainly make everyone more knowledgeable on things, and could even lead to more ideas being shared.
An open culture is a growing culture. When people can avoid segmentation and instead commit to all working together and serving each other, knowledge flows more freely, and growth follows suit.
Interested in all of your thoughts below.
1 Like
You might be interested Jensen Huang’s approach:
They never hear me say something to them that is only for them to know.
There is not one piece of information that I secretly tell the staff that I don’t tell the rest of the company. In that way, our company was designed for agility, for information to flow as quickly as possible, for people to be empowered by that they are able to do, not what they know. That’s the architecture of our company.
This is smart. I would like to adapt this as best I can
Ideas:
- don’t use email for internal discussions - use Issues/PRs/public chat rooms, etc. Then anyone now or in the future can have visibility.
- don’t use 1-1 chat sessions - have topics that anyone can follow or observe as they have interest.
This is mostly how OSS projects work.
Email is where knowledge goes to die …
We use nextcloud internally. Thinking of hosting a Discourse forum on there…
My roommate and I got into a discussion the other day about OpenAI’s value. I made the statement that I think OpenAI is way overvalued and potentially worthless, because I can’t see where their value comes from. My argument was as follows:
An AI company’s value comes from one of three things:
- The model
- The tools
- The compute
So, in the case of ChatGPT, do they have the best model? The best tools? The fastest or most efficient (tokens/sec/$ of energy) compute? Well, I am not sure. But this has to be the source of their value.
My broader point, though, was that all of this value is decimated by the OSS community. ChatGPT may have the best model RIGHT NOW, but in a week there may be an open source model that destroys it. I do not believe ChatGPT has the best tools, and I do believe the OSS community will always win on tools, because we can package a tool however we want. Finally, the compute. ChatGPT is always limited in the profit they can make from this by how cheap it would be for me to either buy my own hardware or host my model on Azure/Linode/etc.
Case in point of this, Jeff Geerling posted this video: https://youtu.be/dQ841Pd6YvQ?si=QElf4XS0YkXFpPuv
ElevenLabs’ entire business model revolved around people using their service. Now I can host it for myself. So now, the value ElevenLabs provides is neither the model, nor the tool, it is the compute.
Where do you think OpenAI’s value comes from? How can being an “AI” company be profitable? Is it a feature or a product?
If I was investing right now (which I am), there are two places I am investing: hosting services (a la Azure, Cloudflare, AWS), and NVIDIA/other AI hardware companies.
An aside: if anyone can figure out how to distribute training compute to millions of devices (smartphones) at a time, I think hardware gets lost too. I’m betting someone in OSS will figure this out.
If your company gets sued, especially by a government regulator, then expect every piece of internal communications to be exposed to the public. Do not write down, in any form, anything illegal or unethical, ever. If you are unsure, then have a voice conversation which is not recorded to discuss.
Lots of industries have data retention policies exactly for this reason, so that when they get sued (not if) that they can easily produce the data for trial discovery to show what happened and who said what.
On the flip side, not the getting-sued-side, one reason things often have to be secret within a company and information isn’t freely shared with other employees is due to the legal structure of contracts. For example, if your company is working with a product from a vendor then that vendor may require that their documentation only be shared with specific people who need to know, as the vendor is trying to control documentation release. This may be due to the vendor being extra cautious, but often it’s because that vendor is a customer of some other vendor and they don’t have the rights themselves to let you share their info internally with others.
Then there’s also human resources rules and regulations. HR often can’t share info about their day to day interactions with employees with anyone except other people in HR and specific employee managers because there’s laws saying so or because the company’s insurance policy says so. Often I think HR is just scared of getting sued, which is a totally legit reason as well.
But then just on the logistics side of things, once you’re beyond about 5 people in a company, if everything everyone did each day was published in some constant feed, it would simply be an overwhelming amount of information where the signal to noise ratio is horrible.
1 Like
It’s just enterprise software, AI from the big name companies will end up with huge contracts from big companies who don’t want, or can’t, bring the tech in-house themselves. Just like how cloud computing in general has gone, sure tech people have hobby projects that use AWS but the vast majority of AWS’s revenue comes from big fat enterprise contracts. I suspect OpenAI and Anthropic will follow suit.
I think it’s really hard to value these AI companies. Are they overvalued based on current financial projections? Yes. Are those financial projections accurate? Who knows. This is very early days in a potentially giant growth tech industry. I personally wouldn’t short sell any of these AI companies with my own money.
I don’t think any of them are scared of open source. They definitely should keep their eyes on the open weight models, many of which are coming out of China, but open weight and open source are very different. There’s lots of open weight models, there’s very few open source ones. In my eyes, open source would mean you publish your entire training datasets and methods under a liberal license (ie: AllenAI do this: Olmo from Ai2) but very few if any of the Chinese models publish their training info.
As far as I can tell, finding GPU compute hardware to rent is extremely hard right now. The big AI players have locked down huge contracts to rent from the big cloud companies or they’re standing up their own data centers. For everyone else, trying to find 8xH200 rigs to rent at a reasonable price is quite difficult. Standing up your own production grade hardware is not cheap, hence the rental demand.
1 Like
@jsnapoli1 By “everyone”, do you mean transparency within a company, or outside the company as well?
@bradfa provided some good examples of the business realities regarding transparency - good information.
I lean toward transparency within an organization when possible. I think systems like Discourse could be used very effectively where most communication is on open channels, and you subscribe/join the Categories that you are interested in, but you have access to everything else when needed. Powerful search/AI capabilities making finding information easy, vs having to go ask people for everything (I should enable AI on this site). Additionally, people who join later have access to history. The key is the flexibility to see what is relevant, decide for yourself what that is, and be able to filter out noise.
Unless you are Apple or Tesla, outside the organization, practically, no one cares. We often think our ideas are pretty special, but if you’ve ever published an OSS project, you quickly learn that unless you are DHH famous, most people are not interested. Reminds me of this quote:
Don’t worry about people stealing your ideas. If your ideas are any good, you’ll have to ram them down people’s throats. – Howard Aiken
When I worked at Xerox we had a formal process in order to release a new component for a machine into manufacturing and to field service (ie: to repair a broken machine). It didn’t matter how big or small the part was, everything followed this procedure. When you wanted to issue a new version of a part then you had to submit your formal declaration to a committee stating:
- What’s wrong with the old one(s)?
- What is the new one? Who do we buy it from? How much does it cost? What are its component parts?
- What do we do with the existing old ones still in inventory in manufacturing but not yet included into product?
- What do we do with the existing old ones in manufacturing which are already installed into product but those products haven’t yet shipped to customers?
- What do we do with the existing old ones stocked in distribution centers for field service? What about units in field service vehicles as urgent spares?
- Do we need to have a way for field service to know that this new one is in a machine?
- Does the machine need to know it has this new one inside itself?
- How much scrap and/or rework charge will this plan incur? Who is paying for it?
- How fast do we need to execute each of these? (ie: is there a new law or a safety concern where we need to rush?)
Each of these questions had to be answered in order to get a new part into the system. The committee had representatives from a variety of internal departments who all had to agree that your plan was good. If your plan passed, then these people would take an execute the plan with their teams.
The amount of data generated by these meetings was immense. Most engineers had no use for this meeting’s data, right up until they did, but then only for very specific reasons. The data was all available for any engineer to look up but unless you wanted to make a change to a part, you just ignored that data as it was just noise to you.
This was a very transparent process and the transparency was amazingly valuable.
1 Like