SLIDE 1
Feedback Loops for Software Delivery
and how to become 6 orders of magnitude more productive
Gojko Adzic | @gojkoadzic | https:/ /gojko.net | Software Craftsmanship London 2019
Feedback Loops for Software Delivery and how to become 6 orders of - - PowerPoint PPT Presentation
Feedback Loops for Software Delivery and how to become 6 orders of - - PowerPoint PPT Presentation
Feedback Loops for Software Delivery and how to become 6 orders of magnitude more productive Gojko Adzic | @gojkoadzic | https:/ /gojko.net | Software Craftsmanship London 2019 Is it working? instapoll.me/992 Studied electrical engineering?
SLIDE 2
SLIDE 3
SLIDE 4
Studied electrical engineering?
instapoll.me/993
SLIDE 5
SLIDE 6
Feedback loops allow us to control
systems with very complex dynamics
using something much much simpler
SLIDE 7
video by EAA museum
SLIDE 8
Glenn H. Curtiss Museum
SLIDE 9
Feedback loops allow us to control
systems with very complex dynamics
stuff we don't fully understand
using something much much simpler
with stuff that fits into our heads
SLIDE 10
SLIDE 11
ConvertFileFunction: Type: AWS::Serverless::Function Properties: DeploymentPreference: Type: Linear10PercentEvery10Minutes Alarms:
- !Ref CheckForLambdaErrors
- !Ref CheckForDropInSales
- !Ref CheckForDropInConversion
Hooks: PreTraffic: !Ref ClearStatistics PostTraffic: !Ref NotifyAdmins
SLIDE 12
The principle of feedback:
Base correcting actions on the difference between desired and actual performance
Feedback Systems, K. Aström, R. M. Murray
SLIDE 13
A £75m “Agile” BBC IT project has evaded scrutiny because managers could make up the benefits as they went along...
https://www.theregister.co.uk/2016/05/10/nao_bbc_projects_agile/
SLIDE 14
A loop without feedback is just RUNNING IN CIRCLES!
SLIDE 15
TDD leads to good design
instapoll.me/996
SLIDE 16
we tried TDD but...
» tests were hard to write » tests are too slow » the architecture/design still ended up bad
SLIDE 17
Scrum's not working...
...we still have all these problems...
SLIDE 18
http:/ /buzz.money .cnn.com/2012/08/02/knight-trading-loss/
SLIDE 19
Feedback without an action is not a loop! It's bad news.
SLIDE 20
Experts are fine with feedback
...
Beginners need actionable advice
SLIDE 21
Drawbacks of feedback loops
» create inherent coupling between parts of a system » can introduce instability if not designed properly » cost and complexity of embedding sensing and actuation
Feedback Systems, K. Aström, R. M. Murray
SLIDE 22
Beware of coupling things through feedback!
https:/ /www.wired.co.uk/article/revolut-trade-unions-labour-fintech-politics-storonsky
SLIDE 23
“People are good at tailoring their behavior to things they are measured against”
Cem Kaner
Don’t Use Bug Counts to Measure Testers, ST&QE 1999, http:/ /kaner.com/pdfs/bugcount .pdf
SLIDE 24
Instability
» measuring the wrong thing » measuring the right thing in a bad way » long loops, delays and oscillations » runaway behaviour
SLIDE 25
Do you run A/B tests in production?
instapoll.me/997
SLIDE 26
SLIDE 27
SLIDE 28
“162 - 480 LOC per staff month”
Capers Jones - Economics of Software Quality
SLIDE 29
https:/ /ieeexplore.ieee.org/document /6127846
SLIDE 30
The value of measuring a variable is usually inversely proportional to how much measurement attention it usually gets
Measurement inversion, Douglas Hubbard - How to measure anything
SLIDE 31
People measure crap that's easy, not stuff that's valuable
SLIDE 32
Do you use automated client-side error tracking (Sentry, Rollbar, crash analytics...)?
instapoll.me/994
SLIDE 33
Answer Responses Not tracking errors 53% Web analytics 15% Sentry 15% TrackJS 4% Rollbar 3%
https:/ /gojko.net /assets/201802-js-research.pdf
SLIDE 34
How frequently do you observe actual users working with your software?
instapoll.me/995
SLIDE 35
https:/ /gojko.net /2016/03/14/most-important-lesson/
SLIDE 36
The faster the feedback, the more intuitive the controls
SLIDE 37
not possible ⇒ expensive
but... does it inform an important decision?
SLIDE 38