Weekly Head Voices #202: I don't even have to miss Dr Topol!

Welcome everyone to this, the two hundred and second edition of the Weekly Head Voices, looking back at the two weeks from Monday July 27 to Sunday August 9 of the year of our simulation-running overlords 2020.

Figure 1: Photo of multiple instances of an important local flower, taken by GOU #1.

Figure 1: Photo of multiple instances of an important local flower, taken by GOU #1.

Shape Up: Stop Running in Circles and Ship Work that Matters

After it having spent a few quality months in my to-do list, I finally got around to reading the Shape Up book by Ryan Singer.

(You can too, by clicking on that link, it’s free!)

This book documents the development methodology that the folks over at 37signals, also the original makers of Ruby on Rails, Basecamp, and most recently the new email re-imagining HEY, have evolved over the years.

I’m not currently using any of those artifacts, but RoR has made (and is probably still making) a significant impact on the technology landscape, and the other products are truly impressive in their own right.

My point is, the 37signals company ships high-impact products like it’s nobody’s business, so one should probably pay attention if they are willing to explain to the rest of us how they are able to perform this feat so consistently.

If you’re involved in software development, or are interested, then this compact book is worth your time.

However, until you can get around to it, I present to you the book’s own bullet-list summary (you’ll find it in the conclusions chapter), where I’ve embellished each bullet with my from-memory explanation.

Shaped versus unshaped work
In Shape Up, each project is a user-visible feature that can be done in 6 weeks or less. Senior devs in the company shape the work, which means that they derisk and write down enough to delineate the work, to specify the parameters, but not so much that it has been pinned down. The team that gets the project assigned is given free reign as to how exactly they fill in the space delineated by the shaping.
Setting appetites instead of estimates
Instead of specifying a feature, and then estimating how much time it will take (feature first, number second), Shape Up suggests rather determining at management level how much appetite / motivation there is for a feature, up to a maximum of 6 weeks, knowing that the feature will be hammered down to fit into the appetite for it (number first, feature second).
Designing at the right level of abstraction
As mentioned above, the shaping process has to reach the Goldilocks level somewhere between abstract and concrete. Enough detail so there are no surprises that could risk the success of the project, but little enough so that the team has the freedom to construct a solution independently and in a more data-driven fashion.
Concepting with breadboards and fat marker sketches
These are practical tricks to keep the detail level under control. For example, during shaping, a fat marker is used to draw UI ideas to prevent the shapers from inking in too much detail.
Making bets with a capped downside (the circuit breaker) and honoring uninterrupted time
This is pretty hardcore. If a project runs over its time (e.g. 6 weeks), it is cancelled, that is, moth-balled. End of story. As a corollary, teams are not to be interrupted during their project cycle.
Choosing the right cycle length (six weeks)
Six weeks is the maximum time a team will get to finish a feature. This seems to be the sweet spot for features that are significant enough to be interesting, but don’t get too big to manage.
A cool-down period between cycles
After every six week cycle, all teams get two weeks of cool-down time. This is to be used for fixing bugs, writing tests, doing spikes, and just generally recovering from the feature development cycle.
Breaking projects apart into scopes
As a team works on a certain feature, the sub-tasks that they formulate as they work, can be naturally grouped into thematic scopes. It’s important that these emergent groupings are closely monitored and updated. They are useful for understanding the feature, and for reporting on progress.
Downhill versus uphill work and communicating about unknowns
The team gives asynchronous feedback on project progress by plotting each scope on a hill chart. To me this is yet another interesting and useful concept from the book. If a scope is somewhere on the uphill part, there are still unknowns that have to be investigated. Once the team thinks there are no more unknowns, a scope can be moved past the top and somewhere onto the downhill section.
Scope hammering to separate must-haves from nice-to-haves
In this case, scope is used in the more traditional software development sense, and scope hammering refers to the action of whittling down features until a project fits into the time allotted to it.

At work, we are currently experimenting with elements of this methodology for a new product we’re working on. I’m curious to see how much of this will work in our setup. It sure looks good on paper!

(The book impressed me further with its chapter containing practical advice on how to implement various levels of Shape Up in organizations of various types and sizes. Super pragmatic!)

The Pygmalion Effect

I’m currently reading the organizational psychologist Adam Grant’s book Give and Take.

In this book he describes the classic study led by Harvard psychologist Robert Rosenthal, where he told the teachers of 18 different classrooms that 20 percent of their children had scored high on a Harvard aptitude test, when in fact they had been randomly chosen.

In spite of being randomly chosen, those children showed consistently higher performance than their peers, even when they were measured years later.

From Grant’s book:

The Harvard test was discerning: when the students took the cognitive ability test a year later, the bloomers [the randomly selected kids, ed.] improved more than the rest of the students. The bloomers gained an average of twelve IQ points, compared with average gains of only eight points for their classmates. The bloomers outgained their peers by roughly fifteen IQ points in first grade and ten IQ points in second grade. Two years later, the bloomers were still outgaining their classmates. The intelligence test was successful in identifying high-potential students: the bloomers got smarter – and at a faster rate than their classmates

and then a paragraph or two later:

Teachers’ beliefs created self-fulfilling prophecies. When teachers believed their students were bloomers, they set high expectations for their success. As a result, the teachers engaged in more supportive behaviors that boosted the students’ confidence and enhanced their learning and development. Teachers communicated more warmly to the bloomers, gave them more challenging assignments, called on them more often, and provided them with more feedback. Many experiments have replicated these effects, showing that teacher expectations are especially important for improving the grades and intelligence test scores of low-achieving students and members of stigmatized minority groups.

Just to reiterate: These kids were randomly selected, but just because their teachers thought that they had scored higher on an aptitude test and hence treated them differently, they started and then kept out-performing their peers for years.

Grant writes that similar experiments have been successfully performed in different learning contexts, and the results were always the same.

This really blows my mind folks. I hope I can keep this knowledge close.

Yet another social media fast because you have to keep on doing them

The Great Social Media Fast of 2019 really was great when it started, but it unfortunately fizzled out a few months later, probably when the end-of-the-year vacation started.

When COVID-19 hit, I had such a need for up-to-the-minute news that my phone soon had twitter almost permanently on position #1 of the app suggestions you get when you swipe down.

Fast-forward another few months, and I was again stuck in TwiFRaF-Yo like it was nobody’s business.

Anyways, this weekend I had some serious discussions with myself, during which I again had to convince myself that the select few jewels of Twitter (Dr Topol, I salute you) and the rest of TwiFRaF-Yo did not even come close to justifying the amount of mental cycles I was spending.

Fortunately I won that debate, and so now I’m again officially on a social media fast, during which I have resolved to read more books, blogs and other long-form content, and to write more notes and blogs.

My brain already feels substantially more roomy and relaxed than it did a mere day ago.

I would be remiss if I didn’t mention one of my favourite College Humor skits in this context:

A not quite satisfactory solution for not having to miss Dr Topol

Dr Topol did briefly complicate my temporary exit.

How was I going to keep up to date with his twitter stream without using the twitter app and getting sucked in?

Unfortunately, my inoreader “supporter” subscription does not include its social media subscription function that does exactly what I need, and the upgrade is $50 per year instead of my current $20.

While I think about this, I’ve activated the IFTTT twitter to email thingy.

This sends me an email for each tweet, which constitutes a whole new problem in itself.

Inoreader satisfaction with a side of retail therapy

… 8 hours pass after this post was published …

Narrator: It became a whole new problem in itself.

Dr Topol can’t stop tweeting admittedly truly interesting health- and covid-related research, but the format of one email per tweet is indeed not going to cut it.

I have deactivated that IFTTT thingy, and upgraded my inoreader subscription to pro.

This is where I read all my blog subscriptions when I make time for that, so having Dr Topol’s feed there as well makes a great deal more sense.

An AI to fall in love (with)

Lex Fridman is another of the internet’s jewels, who is fortunately accessible via long-form podcast.

In episode #106, he talks with Prof. Matt Botvinick, a brilliant neuro-scientist who also employs AI in his research, and clearly likes to think deeply about deep questions.

One of the things I (and apparently many of his interviewees) love about Dr Fridman, is the left-field questions he often poses.

He does this with Botvinick:

But let me ask the over-romanticized question: Do you think we’ll ever engineer an AGI system that we humans would be able to love and that would love us back? So have that level and depth of connection?

Goodness, so many layers.

What do you think?