You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
-**CG Cookie - Reliance on Tutorials**: https://cgcookie.com/community/6394-does-anyone-else-get-the-feeling-that-they-re-reliant-on-tutorials-for-everything-or-just-not-retaining-knowledge-in-general
6
7
-**DEV Community - Identifying Knowledge Gaps**: https://dev.to/bgord/how-do-i-identify-my-knowledge-gaps-and-learn-4mlc
@@ -13,6 +14,7 @@ Content exists for complete beginners ("Hello World") and for experts (advanced
13
14
## The Curse of Knowledge
14
15
15
16
Tutorial authors suffer from the **curse of knowledge**: the inability to remember what it was like not to know something. This causes them to:
17
+
16
18
- Skip steps that seem "obvious" to them
17
19
- Use jargon without explanation
18
20
- Assume prerequisite knowledge without stating it
@@ -21,25 +23,29 @@ Tutorial authors suffer from the **curse of knowledge**: the inability to rememb
21
23
## Key Observations
22
24
23
25
### Tutorials as Peer Communication
26
+
24
27
Many tutorials function like academic papers -- sharing discoveries among professionals who already understand the ecosystem. They're not actually teaching; they're showing off techniques to peers.
25
28
26
29
### The "Crumble" Effect
30
+
27
31
From CG Cookie: "That feeling when you're a beginner, a few months/years in and you've watched tutorials and learnt concepts, but when it comes to making something from scratch you just crumble."
28
32
29
33
This is the exact "rest of the owl" moment -- you can follow along but can't reproduce independently.
30
34
31
35
### Invisible Prerequisites
36
+
32
37
Theoretical knowledge gaps are particularly hard to identify. "It's usually hidden in a talk or a well written article or post." You discover what you didn't know by accident, not by systematic study.
33
38
34
39
### Two Principles for Bridging the Gap
40
+
35
41
1.**Repetition** of important information
36
42
2.**Explicit explanation** of assumed knowledge
37
43
38
44
## Relevance to Plugin Development
39
45
40
46
1.**AI doesn't have the curse of knowledge** -- it can be prompted to explain every intermediate step, state every assumption, and define every term. This is a fundamental advantage.
41
47
2.**A spec-generation plugin should detect and fill assumption gaps** -- when generating a plan, it should make implicit knowledge explicit.
42
-
3.**The "crumble" effect maps to the scaffolding problem** -- users can follow AI-generated code but can't modify or extend it. Better specs would help users understand *why* the code is structured as it is.
48
+
3.**The "crumble" effect maps to the scaffolding problem** -- users can follow AI-generated code but can't modify or extend it. Better specs would help users understand _why_ the code is structured as it is.
43
49
4.**Systematic gap identification is a product feature** -- instead of discovering gaps by accident, a plugin could analyze what the user knows (from their prompt/context) and proactively explain what they'll need to know.
@@ -15,7 +16,7 @@ A development paradigm where well-crafted software requirement specifications se
15
16
16
17
## The Core Workflow
17
18
18
-
1.**Specify**: Share your system idea with an AI agent (the *what* and *why*). Agent generates a detailed specification.
19
+
1.**Specify**: Share your system idea with an AI agent (the _what_ and _why_). Agent generates a detailed specification.
19
20
2.**Plan**: Define the technical approach -- frameworks, tools, languages.
20
21
3.**Tasks**: Break everything into small, structured work packages. Instead of "build authentication," you get "create a user registration endpoint that validates email format."
21
22
4.**Implement**: Agent implements each work package.
@@ -33,6 +34,7 @@ A development paradigm where well-crafted software requirement specifications se
33
34
## SDD vs Vibe Coding
34
35
35
36
SDD is explicitly positioned as the antidote to vibe coding's weaknesses:
37
+
36
38
- Vibe coding: "describe goal, get code back, often looks right but doesn't quite work"
37
39
- SDD: "write complete requirements and technical specs before passing to AI agent"
-**The New Stack - Testing Kiro**: https://thenewstack.io/aws-kiro-testing-an-ai-ide-with-a-spec-driven-approach/
6
7
-**DEV Community - What I Learned Using SDD with Kiro**: https://dev.to/aws-builders/what-i-learned-using-specification-driven-development-with-kiro-pdj
@@ -27,13 +28,15 @@ This prevents "spaghetti code generation" from free-wheeling chat agents.
27
28
## User Experience Reports
28
29
29
30
### Positive
31
+
30
32
- "Spent more time upfront articulating what I wanted to build, but then could step back and let it execute"
31
33
- "The difference between being a hands-on manager versus setting clear expectations and trusting the process"
32
34
- "Kiro did not invent good engineering practices. It made them unavoidable."
33
35
- "It taught users how to think before writing code. That turns out to be the hardest and most valuable part of engineering."
34
36
- Teams report "reducing time to customer value from weeks to days"
35
37
36
38
### Negative
39
+
37
40
- 50 interactions/month on free tier runs out fast
38
41
- Spec overhead is friction for simple tasks
39
42
- Learning curve for EARS notation and formal spec writing
Fixed workflows don't fit all problem sizes. Using heavyweight spec processes for small fixes creates absurd overhead.
38
44
39
45
### Review Burden
46
+
40
47
SDD doesn't eliminate review -- it shifts it from code review to spec review, and the specs can be MORE tedious to review than code.
41
48
42
49
### Instruction Non-Compliance
50
+
43
51
Despite comprehensive specifications, agents frequently ignored instructions or misinterpreted existing code as new specifications, creating duplicates. The spec doesn't guarantee the AI will follow it.
44
52
45
53
### Unclear Target Users
54
+
46
55
Documentation doesn't clarify whether SDD suits small fixes, large features, or requires cross-functional teams.
47
56
48
57
### Semantic Diffusion
58
+
49
59
The term "spec-driven development" is already poorly defined and experiencing semantic diffusion -- "spec" is becoming synonymous with "detailed prompt."
0 commit comments