THEDE TECHNOLOGIES

Eliminating Waste in the
AI Augmented SDLC

Adam Thede

AI Tinkerers Nashville — March 25, 2026

WHERE I LIVE

A terminal emulator is my #1 app. That's where Claude Code runs.

Watch the work.
Find what doesn't add value.
Eliminate it.
Repeat.

Taiichi Ohno, Toyota Production System, 1950s

THE BEFORE — MANUAL /DEVOPS PROCESS

GCP Console Write log queries, parse JSON, count per-URL, per-day
Sentry 8 separate searches — unresolved, new, escalating, regressed, slow, errors, transactions, performance
GitHub Dependabot alerts, open PRs
Jira Search for matching tickets
Manual Cross-reference everything by hand
Manual Write the report

100 minutes

EVERY MORNING

Walk the list on screen:

GCP Console → log queries for 504 timeouts, instance health, Puma worker crashes, thread-holder analysis (which endpoints hog resources)
Sentry → 8 searches: unresolved issues, new issues, escalating, regressed, slow requests, error counts, transaction volume, performance issues (N+1 queries)
GitHub → Dependabot security alerts, open PRs, check if a fix is already in progress
Jira → cross-reference every finding against existing tickets — does a ticket already exist? is someone already working on it?
Then manually connect all the dots and write an action plan

⚡ HIT THE NUMBER: "100 minutes. Every single morning. Before I wrote a line of code."

🖥 LIVE DEMO — flip to terminal, run /devops

17 checks in parallel: 5 GCP, 9 Sentry, 2 GitHub, 1 Jira
Cloud Logging → timeout errors, server health, crashed workers, resource hogs
Sentry → unresolved, new, escalating, slow requests, error counts, traffic, N+1s
Connects the dots: fix in progress? ticket exists? if not, flags it

Tech debt aside (landed well in practice):

Sole engineer → technical debt accumulated
Before: felt insurmountable to profile every controller, every query
After: introduced Claude to the tools → became almost fun
Claude keeps watch → much more effectively tackling tech debt

3-5 minutes. Same coverage. Better accuracy.
Report saved as markdown → indexed and searchable over time

THE NUMBERS — WEEKLY SAVINGS

Skill	What it does	Manual	Auto	Saved/wk
/devops	17 infra checks + cross-reference	100 min	5 min	475 min
/review-pr	Bot comment triage + auto-fix	40 min	5 min	105 min
/wrap-up	Session documentation	48 min	3 min	135 min
/customer-feedback	Feedback logging + dedup	20 min	2 min	18 min
TOTAL				733 min

12.2

HOURS PER WEEK RECOVERED

/devops — 475 min/wk saved

~8 hours saved from one skill I run every morning with coffee

/review-pr — 105 min/wk saved

Just me → rely heavily on bot reviews (Copilot, Code Rabbit, Gemini)
Monitoring bot comments is not fun
Claude reviews them → triage, classify, auto-fix real issues, dismiss bikeshedding
Tied into loop skill → watches PR until bot comments quiet down → "now go look at it"

/wrap-up — 135 min/wk saved

Customer hits tech support → open fresh Claude session → diagnose, research, triage, fix
When done → /wrap-up writes the doc: what, who, why, how we fixed it
Key detail: outputs a CX-friendly message for my team to share with the customer

/customer-feedback — 18 min/wk saved

"You know what would be nice?" → take their exact words → drop next to this skill
Keeps a deduplicated backlog sourced from customers

⚡ PAUSE. SAY THE NUMBER. LET IT LAND.

"At least 12 hours every single week. A day and a half of engineering time, recovered."

IT'S NOT JUST THE DAY JOB

PROJECTS

/dispatch

PARALLEL AI AGENTS

review everything

AGENTS PREPARE — HUMANS DECIDE

THE COMPOUNDING LOOP

Pete Swan

github.com/pdswan

Friday AI meetups
at Switchyards

John Rembold

github.com/jbold

MCP conversation
that started everything

Nathan Gonzalez

github.com/copypastedeveloper

Recommended Ghostty
now #1 app at 55h/mo

"This talk exists because of a Friday morning conversation at Switchyards.
What you do with it tonight is the next iteration of the same loop."

Eliminating Waste in theAI Augmented SDLC

Eliminating Waste in the
AI Augmented SDLC