I’ve written a couple posts recently where I briefly mentioned things that “startups get for free”
Large entities inherently have inefficiencies that smaller companies simply do not have. Getting around this requires a very intentional and uncommon philosophy. The natural tendency is for corporations to become slow hogs, growing department sizes and decreasing their rate of output simulataneously.
On some level, you must hire more. The complexity of the thing you build gets feature rich and features breed complexity. This complexity needs to be managed closely with strong barriers between API surfaces, which requires hiring more people.
So you need to hire more to keep growing, but hiring more is likely to slow down your productivity, per Brook’s Law. Doing this correctly is important.
Recent major tech startups have been obsessing over this struggle, attempting to create supermassive startups. Before the past couple years, this has never been done successfully.
Whether recent companies are doing this well is hard to say. They are certainly seeing gains unseen at this scale, but there is a legitimate work-life balance trade-off which may be hard to sustain in the even longer term. Yet, it seems clear that the progress being made here has allowed companies to accelerate past when they should be able to.
They should have outgrown their own acceleration. It’s why large companies mostly buy small ones. You give up innovating. You observe the inevitability of your predicament and relegate yourself to consuming innovators.
So what is the strategy and what is at the heart of the problem in large organizations? Is it simply communication overhead?
I argue that the productivity slowdown comes fundamentally from a disconnect between the priorities of the company and the individual worker. This disconnect creates job roles which only strengthen this disconnect and workflows which worsen the creative output of the company as whole.
Here’s a thought experiment I’m quite fond of. Imagine the CEO of your company sat down with you asked you precisely what you were doing this week. Do you think they would be happy or upset to hear what you have to say?
A majority of large tech companies, I would say, have such an intense disconnect between the priorities at the VP level and the IC that the CEO would be absolutely furious. Try arguing to high leadership that the 30-person meetings and hours spent writing Jira tickets are a good use of time.
This disconnect is most prevalent in B2C (business to consumer) rather than B2B (business to business). As a coworker once put it, in B2C you are not simply building a product, but making an attempt to change culture.
Netflix does not just have a goal to make a good product, they want to change people’s entertainment habits. Instagram wants you to use Instagram. Uber wants you to prefer ride share to public transit.
These are difficult problems (and morally grey) with difficult to define metrics. What does it really mean to “increase user engagement”? In what manner should it be increased? To what end? For what purpose?
Naturally this puts anyone below VP out of alignment with each other. The goal itself is hard to define, so the work and philosophies of each org become incongruent.
The pyramids of each org filter down into more specialized interpretations of divergent concepts. Eventually you sit at the bottom of the totem pole and can only receive directives from the person placed directly above you.
Your goal is now tightly defined to an end likely not shared or sanctioned by most of the people with power. Getting company-wide appreciation becomes nearly impossible and feature launches become self-congratulatory.
Within this bubble, it’s really difficult to feel good about your work. Unless you are in feverish alignment with your manager, you’ll fall down into the rhythms of thinking “well it’s just a job” and you get your self-actualization from something else. A side project or a hobby, perhaps.
More than being difficult to feel good, the tightly defined org structures create a need for many proxies. Your manager needs to shield you from the rest of the company. As the team size grows, you hire more proxies.
The more proxies, the more communication overhead, the greater the need to start measuring. Sure, measurement is always important to ensure you’re actually doing something, but good measuring is hard and no one is doing it.
Instead, you make up metrics. Bandwidth and productivity are measured by arbitrary numbers like ticket count. You chunk projects in a way to show more output without shifting actual workloads.
You are now enterprise-ready.
In contrast, B2B companies have a bit of an easier time here. Only because their goals are more easily shared and the next steps are obvious. B2C companies can (and sometimes do) operate with the same clarity. It’s just harder.
Still, most B2B companies with their clear vision, fall prey to the same pyramids. Your org structure silos ideas and prevents shared company goals. It seems like the thing you’re meant to do, but there is an alternative.
You might already understand why startups don’t have these issues.
They’re small enough that middle management could not be a thing. Their goals are well understood and equally shared. Job functions are blurry; you help out if you’re able to. There’s a direct communication line between engineering and sales. You iterate quicker.
Startups get these for free.
Large companies easily fall into dangerous pyramids. Startups don’t have enough people for a tower, so you get villages.
There’s no clean analogy for supermassive startups. How can you possibly manage this many people at such a scale?
Honestly, I don’t think a lot of these problems were solvable until very recently.
A lot of the issues that large corporations run into that motivates the pyramid is visibility. You cannot quickly get a gut check on the progress of 1000 people without giving accountability to one person who does the same to their direct reports.
High-trust, high-autonomy companies can get away with this. You just remove productivity tracking as a requirement and allow the work to speak on its own merits. This will greatly reduce your hiring ability, but many companies have found the trade-off worth it.
You remove the hierarchy and allow different job functions to collaborate directly. The company values are readily accessible and impactful work is reliably available for anyone who wants it.
You remove the concept of engineering managers and what you get is a highly collaborative company which can move leagues faster than companies still operating with pyramids.
So is the advice just to hire really really well?
I don’t think this is enough. The companies that are doing this well have been able to scale farther than they should, but they will face the same constraints just at a larger number.
The missing piece is the one we’re currently seeing implemented.
We’re seeing a complete takeover of LLM-based tooling that can do these kinds of analysis in seconds on a massive blurry corpus. You remove the need for middle management and scrum masters.
The other day I heard LLMs being described as fundamentally a data tool - a rich query language on large text resource. Issues like “tracking productivity” become now automatable without requiring the manual data entry we require today. Queries are quickly tweaked - we can change the view without changing the source.
We’re seeing the start of this implementation. For all its ills, the AI craze has produced some legitimately good tooling. It just took us some time to get there.