1
00:00:06,969 --> 00:00:09,117
- Principle number four,
build incrementally

2
00:00:09,117 --> 00:00:11,548
with fast, integrated learning cycles.

3
00:00:11,548 --> 00:00:13,315
Kind of self-obvious isn't it?

4
00:00:13,315 --> 00:00:15,325
Let's talk about it a little bit.

5
00:00:15,325 --> 00:00:17,181
I've, honestly, struggled for years

6
00:00:17,181 --> 00:00:19,348
trying to describe agile and waterfall

7
00:00:19,348 --> 00:00:21,112
in as clean of terms as I could

8
00:00:21,112 --> 00:00:24,202
showing the various benefits
and it's always been hard.

9
00:00:24,202 --> 00:00:26,016
This incremental development thing is

10
00:00:26,016 --> 00:00:27,423
not an easy thing to explain.

11
00:00:27,423 --> 00:00:30,017
But at some point I cam
across, I don't remember when,

12
00:00:30,017 --> 00:00:32,362
I just came back to the
original, fundamental

13
00:00:32,362 --> 00:00:36,670
Scientific Method, and
the Scientific Method says

14
00:00:36,670 --> 00:00:40,124
there's something I don't
know and I wanna learn it,

15
00:00:40,124 --> 00:00:41,068
what do I do?

16
00:00:41,068 --> 00:00:43,165
I put forth a hypothesis, we're gonna hear

17
00:00:43,165 --> 00:00:46,445
that word a lot more later,
I create an experiment

18
00:00:46,445 --> 00:00:49,198
to test the hypothesis,
I run the experiment

19
00:00:49,198 --> 00:00:53,373
and I evaluate the results and
I adjust and I do it again.

20
00:00:53,373 --> 00:00:55,845
That's called, in Deming
and Shewhart's world

21
00:00:55,845 --> 00:00:57,785
that's called plan, do, check, act,

22
00:00:57,785 --> 00:01:01,041
that's the basic, in their
view, process, control,

23
00:01:01,041 --> 00:01:03,978
continuous improvement
cycle, plan, do, check, act.

24
00:01:03,978 --> 00:01:06,743
Now if we think about this a little bit

25
00:01:06,743 --> 00:01:09,062
and we think about the former slide

26
00:01:09,062 --> 00:01:11,454
which showed our waterfall processes

27
00:01:11,454 --> 00:01:15,700
creating a result, how
many times did I plan,

28
00:01:15,700 --> 00:01:18,144
one, how many times did I act,

29
00:01:18,144 --> 00:01:21,729
one, and how many times did I check, one.

30
00:01:21,729 --> 00:01:25,102
It's not that you didn't do a
plan, do check, adjust cycle,

31
00:01:25,102 --> 00:01:26,658
you only did it once.

32
00:01:26,658 --> 00:01:29,654
So you headed down this
path of massive uncertainty

33
00:01:29,654 --> 00:01:32,205
building a system you've
never built before

34
00:01:32,205 --> 00:01:34,021
with one chance to get it right,

35
00:01:34,021 --> 00:01:36,033
I don't think that's a really good plan.

36
00:01:36,033 --> 00:01:38,132
So what we wanna do instead is have lots

37
00:01:38,132 --> 00:01:39,700
of those cycles, right.

38
00:01:39,700 --> 00:01:42,381
Apply fast learning cycles,
we wanna go very quickly

39
00:01:42,381 --> 00:01:45,058
through this process learning as we go.

40
00:01:45,058 --> 00:01:47,251
Lots of little PDCA cycles.

41
00:01:47,251 --> 00:01:50,518
What is a sprint in (inaudible)?

42
00:01:50,518 --> 00:01:53,804
We plan the sprint, we execute the sprint,

43
00:01:53,804 --> 00:01:56,887
we review in retrospect
and we make adjustments

44
00:01:56,887 --> 00:01:59,313
for the next sprint, it's
nothing more than that.

45
00:01:59,313 --> 00:02:01,655
So the easiest way I think
to describe the difference

46
00:02:01,655 --> 00:02:04,907
between traditional development
or stage gated development,

47
00:02:04,907 --> 00:02:07,703
agile development, is
not by the manifesto,

48
00:02:07,703 --> 00:02:10,336
it's how many times you have to plan,

49
00:02:10,336 --> 00:02:12,971
do, check and act, how many
learning cycles do you get.

50
00:02:12,971 --> 00:02:17,737
And by the way, the faster the
cycles the faster learning.

51
00:02:17,737 --> 00:02:20,336
And this does a lot of things,
it improves the efficiency

52
00:02:20,336 --> 00:02:23,090
by decreasing the time
between the action and effect

53
00:02:23,090 --> 00:02:24,450
because it's really quick.

54
00:02:24,450 --> 00:02:26,718
So here's my action and here's my effect

55
00:02:26,718 --> 00:02:29,272
and now I have to
remember what went wrong.

56
00:02:29,272 --> 00:02:32,615
Well guess what, I've got
feedback on that piece right now.

57
00:02:32,615 --> 00:02:35,948
So the feedback is
ready, I coded that thing

58
00:02:35,948 --> 00:02:39,864
just a week ago and now I
see why it doesn't work.

59
00:02:39,864 --> 00:02:41,019
That's really fast.

60
00:02:41,019 --> 00:02:43,819
Versus, holy cow, I think those of you

61
00:02:43,819 --> 00:02:46,295
who write software out there as I used to

62
00:02:46,295 --> 00:02:49,195
know the following paradigm,
you wrote some software,

63
00:02:49,195 --> 00:02:51,173
it's been six months, you open it up,

64
00:02:51,173 --> 00:02:54,199
you look at it and you
go, oh, my goodness,

65
00:02:54,199 --> 00:02:56,586
who wrote this code, blah, blah, blah,

66
00:02:56,586 --> 00:02:59,969
I can't believe, oops,
those are my comments.

67
00:02:59,969 --> 00:03:02,810
You forget your stuff, your
mind moves on to new things

68
00:03:02,810 --> 00:03:05,288
and it's very difficult
to go back to that.

69
00:03:05,288 --> 00:03:08,462
In addition, obviously we
reduce the cost of risk-taking

70
00:03:08,462 --> 00:03:11,045
because right here I can pivot.

71
00:03:11,889 --> 00:03:15,061
I can go, oh, good sprint, bad system,

72
00:03:15,061 --> 00:03:17,061
let's change the design in the system.

73
00:03:17,061 --> 00:03:19,124
This is facilitated by small batch sizes.

74
00:03:19,124 --> 00:03:21,387
Why, because I need
just the amount of work

75
00:03:21,387 --> 00:03:25,102
I can do this batch, I'll just
call a little back log here,

76
00:03:25,102 --> 00:03:27,733
in that period, I don't
need this gigantic thing.

77
00:03:27,733 --> 00:03:29,181
And if I only have this gigantic thing

78
00:03:29,181 --> 00:03:30,629
I have to bust it up.

79
00:03:30,629 --> 00:03:33,021
And if I'm being held hostage to proving

80
00:03:33,021 --> 00:03:36,025
everything on that
gigantic requirements list

81
00:03:36,025 --> 00:03:38,705
over time, then I don't
have a small batch size

82
00:03:38,705 --> 00:03:39,739
that I can work with.

83
00:03:39,739 --> 00:03:42,134
So we need small batch sizes
to move it through the system.

84
00:03:42,134 --> 00:03:43,987
And that requires an increased investment

85
00:03:43,987 --> 00:03:45,219
in development environment.

86
00:03:45,219 --> 00:03:47,115
I saw a talk by Jez the other day

87
00:03:47,115 --> 00:03:49,092
where he talked about the HP case study

88
00:03:49,092 --> 00:03:53,545
that's online building
printers applying DevOps

89
00:03:53,545 --> 00:03:56,719
talking about the fact that
they dramatically reduced

90
00:03:56,719 --> 00:04:00,119
the nonproductive time
overall, I don't remember

91
00:04:00,119 --> 00:04:02,636
the exact numbers, but one net effect

92
00:04:02,636 --> 00:04:05,113
was a lot of the nonproductive time

93
00:04:05,113 --> 00:04:07,215
was now invested in test automation.

94
00:04:07,215 --> 00:04:09,232
Leaving, I think, four or five times

95
00:04:09,232 --> 00:04:11,206
the amount for actual
coding and innovation.

96
00:04:11,206 --> 00:04:12,563
So we may have to do that.

97
00:04:12,563 --> 00:04:14,328
We're gonna have to change the way

98
00:04:14,328 --> 00:04:15,765
our investment profile works.

99
00:04:15,765 --> 00:04:18,203
But with that automation,
the shorter the cycles,

100
00:04:18,203 --> 00:04:19,896
the faster the learning.

101
00:04:19,896 --> 00:04:23,115
Fast learning cycles,
plan, do, check, act.

102
00:04:23,115 --> 00:04:26,092
One of my favorite reads and
strongly recommend to you

103
00:04:26,092 --> 00:04:27,628
is Dantar Oosterwal's The Lean Machine,

104
00:04:27,628 --> 00:04:30,311
it's the story of
Harley-Davidson back in the 90s,

105
00:04:30,311 --> 00:04:31,640
you're probably not old enough,

106
00:04:31,640 --> 00:04:33,448
but I remember riding a Harley in the 70s

107
00:04:33,448 --> 00:04:36,086
and they were cool, they were
the coolest machines ever.

108
00:04:36,086 --> 00:04:37,568
If you've never ridden a Harley,

109
00:04:37,568 --> 00:04:39,492
you almost have to, it's
a really cool thing to do.

110
00:04:39,492 --> 00:04:41,388
And then the Honda Gold
Wing Coupe came along

111
00:04:41,388 --> 00:04:43,041
and all the sudden it's very different,

112
00:04:43,041 --> 00:04:45,098
all the sudden there's this really fast,

113
00:04:45,098 --> 00:04:47,659
much more reliable machine
and Harley went into trouble.

114
00:04:47,659 --> 00:04:49,345
So they decided to do adopt lean

115
00:04:49,345 --> 00:04:52,512
and they were coached by Dr. Alan Ward

116
00:04:53,553 --> 00:04:56,353
was their mentor and you'll
see that in the book there

117
00:04:56,353 --> 00:04:58,043
and they went through the process

118
00:04:58,043 --> 00:05:00,022
and Dantar has a lot of learning in there

119
00:05:00,022 --> 00:05:02,203
that I think are synthesized so well.

120
00:05:02,203 --> 00:05:04,505
One of them is that, "Product development

121
00:05:04,505 --> 00:05:06,940
"is a process of converting
uncertainty to knowledge."

122
00:05:06,940 --> 00:05:09,690
Oh, yes, that's it, I don't know.

123
00:05:11,242 --> 00:05:14,909
And having specs helped
me know what I think

124
00:05:15,766 --> 00:05:17,869
I need to do, but I don't really know.

125
00:05:17,869 --> 00:05:19,306
So it's okay not to know.

126
00:05:19,306 --> 00:05:21,855
So in agile and lean thinking,
it's okay not to know,

127
00:05:21,855 --> 00:05:23,506
we're gonna convert
uncertainty into knowledge

128
00:05:23,506 --> 00:05:25,321
and those are controlled
by integration points.

129
00:05:25,321 --> 00:05:27,300
So we come back to the PDCA cycle

130
00:05:27,300 --> 00:05:28,873
and what do we see?

131
00:05:28,873 --> 00:05:31,102
We see, well, if this
is just, for example,

132
00:05:31,102 --> 00:05:34,396
I'm working on the engine
and I'm migrating the engine

133
00:05:34,396 --> 00:05:37,235
more quickly here,
there's a new carburetor

134
00:05:37,235 --> 00:05:38,723
and I'm getting feedback by that,

135
00:05:38,723 --> 00:05:40,084
my carburetor's working pretty good,

136
00:05:40,084 --> 00:05:42,527
seems like the engine runs, but the engine

137
00:05:42,527 --> 00:05:45,282
hasn't been hooked up to
the new drive train yet.

138
00:05:45,282 --> 00:05:47,056
So I've got some new
horsepower, can I handle it,

139
00:05:47,056 --> 00:05:48,704
I don't know 'cause all I've done so far

140
00:05:48,704 --> 00:05:50,439
is prove that the new carburetor works.

141
00:05:50,439 --> 00:05:53,408
And by the way, does the
engine and the drive train

142
00:05:53,408 --> 00:05:57,575
and the frame all hold together
to make a great Harley?

143
00:05:59,372 --> 00:06:00,857
Well that's a bigger cycle.

144
00:06:00,857 --> 00:06:05,024
So what you see in safe is
based upon this principle

145
00:06:05,882 --> 00:06:09,299
of synchronous, harmonic learning cycles.

146
00:06:11,124 --> 00:06:13,433
These cycles are the same,
these cycles are the same

147
00:06:13,433 --> 00:06:15,944
and the slowest one is what controls

148
00:06:15,944 --> 00:06:18,079
the overall rate of learning.

149
00:06:18,079 --> 00:06:21,875
Development can proceed no
faster than the slowest point.

150
00:06:21,875 --> 00:06:24,212
So I don't really know
if I have a full Harley.

151
00:06:24,212 --> 00:06:26,271
Now I don't really know if
I have a full carburetor

152
00:06:26,271 --> 00:06:28,256
til here, so imagine all the experiments

153
00:06:28,256 --> 00:06:29,990
you could do in that time frame.

154
00:06:29,990 --> 00:06:33,452
Important learning, nested PDCA cycles.

155
00:06:33,452 --> 00:06:34,853
I really wanna think about the problem.

156
00:06:34,853 --> 00:06:37,618
And you see that in safe,
you see the team iterating,

157
00:06:37,618 --> 00:06:39,963
you see they are using those together

158
00:06:39,963 --> 00:06:40,998
to integrate the system.

159
00:06:40,998 --> 00:06:43,396
You see the PI cycle so that the arts

160
00:06:43,396 --> 00:06:45,722
can come together with more to do.

161
00:06:45,722 --> 00:06:49,467
What he notes here is
what's built into safe.

162
00:06:49,467 --> 00:06:52,649
Improvement comes through
synchronization of design loops

163
00:06:52,649 --> 00:06:54,790
and faster learning cycles.

164
00:06:54,790 --> 00:06:56,187
That's how you get better.

165
00:06:56,187 --> 00:06:58,667
Synchronized the loops, shorten them.

166
00:06:58,667 --> 00:07:00,477
So if you're running three week sprints,

167
00:07:00,477 --> 00:07:02,575
run two week sprints, if you're
running two week sprints,

168
00:07:02,575 --> 00:07:04,592
think about running
shorter sprints than that.

169
00:07:04,592 --> 00:07:06,779
Really, really fun learnings
from the Harley-Davidson story,

170
00:07:06,779 --> 00:07:08,935
it's also a really fun read.

171
00:07:08,935 --> 00:07:10,996
What's the net result of that for risk?

172
00:07:10,996 --> 00:07:12,769
People with a traditional perspective say,

173
00:07:12,769 --> 00:07:15,643
where's the risk mitigation plan in safe?

174
00:07:15,643 --> 00:07:17,329
Where do yo manage risk?

175
00:07:17,329 --> 00:07:19,636
The risk mitigation plan in safe

176
00:07:19,636 --> 00:07:21,908
is the fact that integration
points reduce risk,

177
00:07:21,908 --> 00:07:23,971
objective evaluation working systems

178
00:07:23,971 --> 00:07:25,490
that we'll talk about in a second

179
00:07:25,490 --> 00:07:27,472
is the thing we use to reduce risk.

180
00:07:27,472 --> 00:07:29,662
Now, what Harley also points out

181
00:07:29,662 --> 00:07:31,314
is that they always thought
that they were building

182
00:07:31,314 --> 00:07:34,451
good Harleys and sometimes
they would go out here

183
00:07:34,451 --> 00:07:36,026
and crash and burn,
well maybe not literally

184
00:07:36,026 --> 00:07:38,216
crash and burn, but not
have a good release.

185
00:07:38,216 --> 00:07:39,827
And they're going, well, what is this?

186
00:07:39,827 --> 00:07:41,475
We thought we were building a Harley

187
00:07:41,475 --> 00:07:43,454
and we did build it, it
just didn't really work.

188
00:07:43,454 --> 00:07:46,175
It crashed and burned at the end.

189
00:07:46,175 --> 00:07:48,033
They call that false positive feasibility.

190
00:07:48,033 --> 00:07:51,495
That's managers worst nightmare.

191
00:07:51,495 --> 00:07:53,727
I think it's okay, but I'm not.

192
00:07:53,727 --> 00:07:55,779
That false positive feasibility

193
00:07:55,779 --> 00:07:58,748
represents the risk that remains inside

194
00:07:58,748 --> 00:08:00,969
the waterfall project until such time

195
00:08:00,969 --> 00:08:03,686
as you've done a full integration.

196
00:08:03,686 --> 00:08:05,749
You don't know until
the integration happens.

197
00:08:05,749 --> 00:08:09,906
So that risk remains, all that
risk in the project remains.

198
00:08:09,906 --> 00:08:12,132
Well we start out agile development

199
00:08:12,132 --> 00:08:14,081
with the same degree of risk for sure,

200
00:08:14,081 --> 00:08:15,852
but we move very quickly down this path.

201
00:08:15,852 --> 00:08:18,910
Each iteration, sprint, PI,
whatever you wanna call it,

202
00:08:18,910 --> 00:08:21,090
whatever that frequency is reduces risk.

203
00:08:21,090 --> 00:08:24,440
So when we get to the
end we're truly done.

204
00:08:24,440 --> 00:08:27,200
This is the boring release
of your next CRM system,

205
00:08:27,200 --> 00:08:29,184
the next big release of our framework

206
00:08:29,184 --> 00:08:32,830
on top of the latest version
of WordPress, it's boring.

207
00:08:32,830 --> 00:08:35,020
Why, because we found all the problems

208
00:08:35,020 --> 00:08:36,586
and addressed them integrally.

209
00:08:36,586 --> 00:08:38,564
So those are absolutely critical ways

210
00:08:38,564 --> 00:08:40,251
to think about mitigation of risk.

211
00:08:40,251 --> 00:08:41,743
So if you're working with folks

212
00:08:41,743 --> 00:08:43,106
that believe that risk is an issue,

213
00:08:43,106 --> 00:08:44,799
and they should, we
have an answer for that

214
00:08:44,799 --> 00:08:47,068
which is, yeah, we're
gonna address the risk

215
00:08:47,068 --> 00:08:50,863
right upfront by basically
integration points

216
00:08:50,863 --> 00:08:54,197
and objective evaluation
of working systems,

217
00:08:54,197 --> 00:08:56,537
which is principal number five.