Accelerate - The Science of Lean Software and DevOps

A #book by Nicole Forsgren, Jez Humble and Gene Kim (forwards by Martin Fowler and Courtney Kissler)

Recommended to me by Joe Peacock after DDD workshop slides

Just about from commit to prod, not the entire software development lifecycle

learning organization

failure leads to inquiry

Five categories of key capabilities

  1. continuous delivery
  2. architecture
  3. product and process
  4. lean management and monitoring
  5. culture

definition needed: waterfall methodology

improvement is possible for everyone

snowball sampling

software delivery can be defined and measured in a stastically meaningful way, software delivery impacts organizational performance

def needed capabilities model

before you can measure success you have to define 'good'

To take a brief assessment, use the DORA quick check

(though cf Goodhart's Law)

The authors go beyond correlation and into prediction (not yet causation) (though apparently change failure rate isn't part of their construct: it didn't pass validity testing and reliability testing. it does predict success)

There are three parts, which go from more implicit to more explicit

  • basic assumptions: things you just know
  • values: they might be discussed
  • artifacts: things written/otherwise observable

An organizational culture based on power. Information is hidden for personal gain. The "worst" of the types of organizational cultures.

An organizational culture focused on process. There are battles over "turf".

bureaucracy isn't inherently bad

Also known as "generative culture". or is it? are you conflating terms?

An organizational culture focused on the mission. What matters is accomplishing goals.

The "best" of the types of organizational cultures.

Is this simplistic? I'm not sure how much it reflects reality and how much it's "do these things to be better". Synacor was almost certainly a bureaucratic culture, though.

Ron Westrum says organizational culture predicts performance

Of note to me, and I think the most realistic (as opposed to the "messengers shot/messengers neglected/messengers trained" example of pathological->bureaucratic->learning) is the response to failure: failure leads to scapegoating, failure leads to justice, failure leads to inquiry. Justice can be tempting[1], but as someone who has failed and been met with inquiry, it feels really good.

in pathological and bureaucratic cultures, measurement is used as a form of control, and people hide information that challenges existing rules, strategies, and power structures. As Deming said, "whenever there is fear, you get the wrong numbers."
-- Humble et al. 2014, p. 56

Essentially, Goodhart's Law.

To change culture don't change how people think, change what they do.

Is this correct? Or is enforcing agile development/etc on people who don't want it going to backfire?

Good information

  1. Provides answers to the questions the receiver needs answered
  2. Is timely
  3. Is presented so it can be effectively used

Likert scale questions should be strongly worded

If you have unplanned work or rework, it means the initial job wasn't good enough.

A concept created by John Seddon[2]: the demand for work (unplanned work or rework) caused by the failure to do it right the first time[1:1].

This is the (bad) counterpart to value demand.

  1. Failure Demand – Vanguard Consulting Ltd ↩︎ ↩︎

  2. I Want You to Cheat, 1992 ↩︎

On this page
Pages mentioning this page
Domain Driven Design Modeling Process - From Problem to Solution Space
Tech