The disciplines that come after DevOps — and why each of them exists because the previous one succeeded.
~5,400 words across 4 notes. Built August 2026. [reference].
The one idea: Waterfall, Agile, DevOps, SRE, platform engineering and DevSecOps are not competing philosophies or a seniority ladder. Each solved the bottleneck the previous one created by working. Optimise any stage of a pipeline and the constraint relocates — that’s the whole sequence.
Why this exists
devops/ had eleven sections covering the tools — pipelines, Terraform, Kubernetes, Prometheus — and devops-reference had two paragraphs on the disciplines that decide how those tools get used. The vault knew what an error budget was (observability 01 covers SLI/SLO/SLA properly) but had zero occurrences of “toil”, nothing on on-call design, incident command or blameless postmortems, one passing mention of Internal Developer Platforms, and no note on DevSecOps at all.
So: the vocabulary of the roles the vault is preparing you to hold, which existed everywhere in job ads and nowhere in these notes.
Reading order
Read 01 first — the other three are its branches.
- How Delivery Practice Evolved — [Beginner] — five transitions, each fixing the last one’s bottleneck. The spine of the folder
- Site Reliability Engineering — [Intermediate] — toil and the 50% cap, on-call design, incident command, blameless postmortems, capacity planning, DORA’s four
- Platform Engineering — [Intermediate] — the two ways a DevOps team breaks, the IDP, golden paths not golden cages, and cognitive load as the actual thesis
- DevSecOps — [Intermediate] — what runs at which pipeline stage, SAST vs SCA vs DAST, SBOMs, policy as code, and the alert fatigue that kills most rollouts
The things worth carrying
- Each era created the next era’s problem by succeeding. Agile made dev fast → ops became the wall. DevOps fixed the wall → nobody owned production. SRE fixed that → the DevOps team didn’t scale → 01
- They coexist. “We’ve moved past DevOps” is nearly always a misreading — the platform is built by people doing DevOps work → 01
- The error budget turns a values argument into arithmetic. Nobody debates reliability vs features; the number decides → 01
- Toil is automatable manual work that scales with the service — distinct from overhead, and capped at 50% with enforcement. Drop the enforcement and SRE becomes a renamed sysadmin role → 02
- Every noisy alert makes the system less reliable. Deleting alerts is real reliability work → 02
- Mitigate first, diagnose second. The rollback is reversible; live debugging on a burning system isn’t → 02
- Blameless isn’t kindness, it’s evidence collection. Punish the name and you lose the detail you needed → 02
- A central DevOps team becomes the wall it was created to remove — and embedding engineers instead just trades the queue for twelve incompatible setups → 03
- Golden paths, not golden cages. Mandatory means teams route around it invisibly → 03
- Measure users’ outcomes, not the platform’s output. Time-to-first-deploy, not modules shipped → 03
- If your platform must be understood in addition to the tools underneath, it made things worse → 03
- SCA is the highest-value scanner for most teams — most of your code is code you didn’t write → 04
- A scanner nobody acts on is worse than none — it converts a known gap into false confidence → 04
- A leaked secret isn’t fixed by deleting the commit. Rotate first, clean history second → 04
Where this connects
| 07-observability/ | The prerequisite. SLI/SLO/SLA and error budgets live there; note 02 deliberately doesn’t repeat them |
| 05-ci-cd/ | The pipeline all four notes assume |
| GitOps | The deploy mechanism most IDPs sit on |
| cybersecurity | Note 04’s other half, at depth |
| the engineering roles | What each of these is like to hold as a job |
| availability | Where the nines come from |
The honest note
[reference], and unusually so — three of these four are organisational disciplines, and you cannot validate an organisational discipline by reading about it or by building a toy. A one-person project has no dev/ops wall, no ticket queue, no on-call rotation and no cognitive-load problem. The conditions that make these ideas necessary are conditions of scale.
What would close the gap, honestly ranked:
- Define one real SLO on something you actually run — including the bit everyone skips: decide in advance what you’ll stop doing when the budget is gone. Note 02 is theatre until that decision is real
- Write a postmortem for a failure you caused. Blameless is easy when it’s someone else. It’s a genuinely different exercise on your own mistake, and it’s the one worth doing
- Track a week of your own work as toil / engineering / overhead. Cheap, and usually uncomfortable
- Add SCA to one repo — Dependabot or Trivy, ten minutes. Then actually resolve what it finds, which is the part that teaches you why baselining exists
- Add secret scanning with a pre-commit hook, then deliberately try to commit a fake AWS key and watch it fail
- Build the smallest possible golden path — one
create-servicescript producing a repo with CI, a Dockerfile and a dashboard already wired. Note 03 at a scale of one, and it still teaches the trade - The reading: the Google SRE Book and SRE Workbook (free); Accelerate (Forsgren, Humble, Kim) for the DORA evidence; Team Topologies (Skelton & Pais) for the cognitive-load argument note 03 rests on
What’s missing: chaos engineering in any depth (it’s in devops-reference and distributed systems testing), FinOps and cost as a reliability constraint, progressive delivery specifics, compliance frameworks (SOC 2, ISO 27001 — see GRC), and anything about the politics of actually introducing these to an organisation that hasn’t asked for them, which is the hardest part in practice.
Related
- DevOps — the folder this closes
- 00-the-physical-layer/ — the other end of the same course
- Interview Prep Index
- Build Plan
Software engineering
Infrastructure & security
Computer science
Hardware & engineering