Halvar Flake Profile picture
May 4 15 tweets 3 min read Read on X
Waiting for a delayed flight, some idle thoughts on what Google needs to fix itself:

1) Find a replacement for Sundar. This person needs to be both able if articulating a coherent vision for Google as a company, *and* inspire great engineers to want to work on great problems.
2) Google is a strange dysfunctional conglomerate. Search/Ads is/was one of the best businesses ever: Extremely high margin, network effects etc. - at the price of requiring heavy R&D for scale, search quality etc. -- a lot of early Google culture was built for such a business.
Many other businesses at Google (devices, cloud, enterprise docs) are very different: lower margins, more sales effort, tight qualitative customer feedback needed etc.

Neither early Googles culture nor pay scales are well-adapted.
Google has largely killed old Google culture as a result, and the cash cow will eventually suffer. So a key step to returning Google back to health is splitting up the conglomerate.

Cloud needs to be split off, possibly devices, possibly YT or Android. The exact shape is TBD.
Several fragments of Google would be significantly sized public companies in their own right.

3) Google has a huge amount of Parkinson's Law growth and empire-building to undo. The fastest way for mediocre people to get big salaries in the period 2011-2022 was to quickly grow...
... a small team into big orgs of 50-200 people. 2-4m TC as the carrot. Headcount and org size ballooned. Search/Ads success subsidized growth that was fat, benign tumors, or cancerous, and hiring quality suffered during the pandemic hiring acceleration.
This was a result of the incentive structure. People got promoted for building big systems where small would have worked, and big teams when small would have worked.
Undoing this is super difficult and super painful. But if there is one lesson from history: Cut once. Cut deeply. Then let new branches sprout carefully.

The rolling layoffs that currently happen are crazy. Demoralizing, and not dealing with the actual rot.

Most likely a ...
... lot of headcount reduction is needed. This might be more easily done *after* splitting the conglomerate, as cross-subsidies from search will become more visible.

4) A new CEOs role is to greatly diminish their role by breaking up the conglomerate and then leading only...
...search/ads, and finding strong CEOs for the fragments.

5) Search/Ads needs to restore the culture that it's the #1 place where people want to go to "make the world's information universally accessible". It needs to become the dream destination for technically ambitious ...
... scientists again. This requires clarity of vision, and the sort of audacity that Larry articulated as CEO: He wanted Google engineers to work right on the boundary of the possible.
6) Google somehow infected itself with Microsoft-internal-warfare-disease: When a company has few external threats, it can spin into a strange mode where most effort is on internal strife. Observing the frequency of strange reorgs with negative product or biz impact because ...
... some executive wanted a shiny piece of biz in their org/portfolio, Google has its own share of a similar disease. The loss of internal transparency exacerbates the issue. Whoever takes over needs to identify who the drivers for internal corporate warfare are, and take...
... an axe and a zero tolerance for bullshit approach to the problem.
@threadreaderapp unroll

• • •

Missing some Tweet in this thread? You can try to force a refresh
 

Keep Current with Halvar Flake

Halvar Flake Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!

PDF

Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

More from @halvarflake

Dec 10, 2023
@bgurley @martin_casado Ok. Warning, this will be long, and interrupted by a flight.
@bgurley @martin_casado 1) Cultural factors: When I was growing up in the 90s, there was significant uncertainty in the labor market, and one way to achieve economic security was seeking a government job. In many European countries, running a limited liability construct into insolvency effectively...
@bgurley @martin_casado ...bans you from running another one in the foreseeable future. The mentality of "start a company in your 20s, and if you fail, you can either try again or get a job" wasn't a thing. So we are operating from a risk-averse base, due to a labor market with then-sluggish...
Read 29 tweets
Nov 2, 2023
I wish I had more time to chime into the AI doom debate, but here a very quick thread:

1) The one thing all AI doomers seem to assume is that almost all engineering problems can be solved by thinking, vs. experimentation.
2) Humanity has seen multiple individuals of ...
... vastly higher IQ than other humans. We're 8bn people, IQ stddev is 15, there's bound to be a few people of IQ 175-190. None of them have proven to be particularly dangerous. Superhuman AI does not mean infinite AI.
3) There's no reason to believe the "intelligence explosion" they seem to believe in can actually happen. Pretty much all processes are self-limiting; it's hard to make deliberate exponential chain reactions happen.
4) For all we know, Orcas may be vastly superhuman, ...
Read 11 tweets
Sep 13, 2023
One thing I have learnt over the last years is that - while I am technically pretty solid - I am surprisingly good at *product*. It's a strange thing to realize as a pretty technical mathematician.

This thread describes what I consider "common-sense product design", because ...
... it turns out that common sense is not all that common.

Ok, so you want to design a good product. Here are the steps:

1. Create a target demographic / user and buyer persona. This comes absolutely first.
2. Identify a few people that match this description. Get to know them.
3. Listen to them, and try to elicit things that make their lives bad, and their incentive structures. Let. Them. Talk.
4. Think about what they complained about, and how you could make their lives better. Come up with a hypothesis.
5. Talk with them about your hypothesis.
Read 10 tweets
May 5, 2023
There's a lot of microservices hate, but there are also terrible balls of yarn. It reminds me that many orgs are not good at engineering.

A few rules that have served me well:

1) architect your software, have a diagram.
...
2) centralize responsibilities in the diagram.
3) not every box in that diagram needs to be a service, some boxes should be shared libraries
4) conversely, not every box should be a shared library
...
5) There are two primary reasons to make a box a service: you need parallelism for a box that one machine cannot do, or the box is used by many other boxes and manages state. (Plz reply if you know other good reasons, my list is incomplete).
Read 6 tweets
Feb 5, 2023
A thread about family culture and how value systems survive and get transmitted implicitly:

My parents were deeply pacifist, through personal experience of WW2. They were involved in the creation of the German Green party, which had roots in the peace movement.

...
At the same time, there was a deep undercurrent of duty, self-sacrifice, serving the state/the greater good, "critical obedience" (kritischer Gehorsam), and holding oneself to extremely (perhaps inhumanely) high standards of personal integrity.

...
As a child, you are immersed in these things, and they are never actively spoken about, so you absorb without noticing.

As a teenager, casting around for value systems for myself, I read the Hagakure, and it resonated deeply with me. I adopted many parts from it as ...
Read 8 tweets
Sep 3, 2022
After experimenting with stable diffusion a little bit, I was trying to make sense of the things I observed, and spend a few minutes spellunking around in the 12m images that one can explore at laion-aesthetic.datasette.io/laion-aestheti….

Observation thread:
1) I was puzzled why particular artists that are (in my eyes) relatively low profile generate significant improvements in output quality when added to the prompt, while other prominent artists in the prompt generate poor results. The training set explains some of it: The term ...
... "manara" appears in 38 out of 12m pictures, leading us to estimate that there's about 1900 such pictures in the training set, wheras the term "artgerm" appears in 967 pictures, so an estimated 48k pictures were in the training set. Adding "artgerm" to the prompt then ...
Read 7 tweets

Did Thread Reader help you today?

Support us! We are indie developers!


This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Don't want to be a Premium member but still want to support us?

Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal

Or Donate anonymously using crypto!

Ethereum

0xfe58350B80634f60Fa6Dc149a72b4DFbc17D341E copy

Bitcoin

3ATGMxNzCUFzxpMCHL5sWSt4DVtS8UqXpi copy

Thank you for your support!

Follow Us!

:(