Add the inverted coding paradigm essay - #580
Merged
Merged
Conversation
Draft of "Inverted Coding Paradigm for AI-Assisted SWE",
arguing that programming should move to asynchronous and
overnight agent loops so that business hours go to the
human review gates, cross-functional collaboration, and
spec work that agents cannot do unattended.
Published at /posts/2026/08/invert-coding-sdlc/ with an
"ai-sdlc" redirect.
Cites Mazur on peer instruction, Bergmann and Sams on the
flipped classroom, Khan on popularizing it at scale, and
Orosz on loop engineering. Further citations are still
being sourced, so the argument in the Gates and Specs
sections is expected to change.
Verify with:
npx prettier --check --print-width 60 \
--trailing-comma=none --prose-wrap always \
_posts/2026-08-11-invert-coding-sdlc.md
✅ Deploy Preview for p13i ready!
To edit notification comments on pull requests, go to your Netlify project configuration. |
The bare [1] through [4] markers had no matching link
definitions, so they rendered as literal bracketed text and
the References list sat unconnected to the body.
Convert them to kramdown footnotes with named slugs, the
same convention as _posts/2022-02-02-friends.md. kramdown
now numbers the markers in order of first appearance,
places the reference list at the end of the page, and adds
a back-link from each entry to its citation site. Named
slugs mean a later citation can be inserted without
renumbering.
Also correct the gate count: the section said "about three"
above a four-item list and "these four types of gates"
below it.
Verify with:
npx prettier --check --print-width 60 \
--trailing-comma=none --prose-wrap always \
_posts/2026-08-11-invert-coding-sdlc.md
Editorial pass answering four independent reviews.
Claims that could not be defended are bounded. Code review
no longer becomes "cursory" after agent pre-review; it moves
up a level, to architecture, security, and whether the
implementation matches intent. Tests no longer imply agents
can be trusted "at human quality"; passing tests prove the
code satisfies the tests, which makes the test matrix the
artifact worth reviewing.
The Specs section now says what actually changed. Spec-driven
generation from OpenAPI and Protobuf always stopped at the
contract surface, emitting models, clients, and server stubs
with empty method bodies. Agents draft the bodies. That
boundary, not input format, is the novel claim.
Adds the review-throughput argument the essay was missing,
including self-reviewable PRs that carry their own evidence
and order the diff by risk.
Replaces assertion with measurement in four places, drawn
from my own runs: the builder census, unattended duration,
review load per overnight batch, and the share of pull
requests that never merged. Each carries a footnote naming
what was measured.
Also converts the four existing citations to kramdown
footnotes with named slugs, adds three more, and fixes six
grammar defects.
Verify with:
npx prettier --check --print-width 60 \
--trailing-comma=none --prose-wrap always \
_posts/2026-08-11-invert-coding-sdlc.md
The share was measured across pull requests in my personal projects over the past few months, not over a single set of overnight builder runs. The sentence now names that population, matching its footnote.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Adds one post:
_posts/2026-08-11-invert-coding-sdlc.md.Nothing else is in this branch.
What it argues
Programming should move to asynchronous and overnight agent
loops, so that business hours go to the work agents cannot do
unattended: the human review gates, cross-functional
collaboration, and the specs and test matrices agents build
against.
The framing borrows the inverted classroom. Mazur moved
information transfer out of the classroom and information
assimilation into it; the essay proposes the same inversion
for code production and code judgment.
Publishing details
/posts/2026/08/invert-coding-sdlc/redirect_from: "ai-sdlc"prettier --print-width 60 --trailing-comma=none --prose-wrap alwaysState of the draft
This is a working draft opened deliberately in public, not a
finished piece. Expect the argument and the citations to move.
Currently cited:
Still to source:
likely from Epoch AI or METR.
claim about agent thrash costing wall-clock rather than
correctness.
Known open items
"about three" gates, then lists four, then refers to "these
four types of gates". The count needs to settle.
"manual validate", and "leads no where".
Verify
Not verified: the page has not been built through Jekyll
locally, because this machine cannot resolve the bundler
version pinned in
Gemfile.lock.Written by Claude