Blog - Konrad Kowalski (rootsher)
pinned
- series · 10 parts
Part II: Standardizing AI in the SDLC Across the Organization
From a single managed agent to a shared, controlled platform for teams. The series shows what has to move out of the agent's surroundings and into shared…
- series · 9 parts
Part I: From Prompts to Agentic SDLC
This series moves from copying prompts between chat and IDE to agents that work in the repository, use context, tools, workflows, external orchestration and…
- series · 6 parts
Deployment Strategies: Why They Exist
Deployment strategies did not appear to make pipelines look serious. They appeared because replacing a version became a moment of risk.
- series · 5 parts
Recovery from the Ground Up: Rollback, PITR, and Returning After Failure
Recovery does not start with a large disaster recovery plan. It starts with a simple question: what exactly do we need to return to, and how much can we lose?
series
- Gateway API Without the Magic4/4
- Testing Through Layers: From Code to Infrastructure9/9
- Ways of Talking to a Backend: From HTML Forms to GraphQL6/6
- Realtime in the Browser: From Polling to WebTransport7/7
- The Layers of HTTP in the Browser: From fetch() to QUIC9/9
- Frontend Rendering: from server-side pages to hybrid rendering17/17
- CPU in Kubernetes from the kernel up12/12
- Kubernetes from the Inside: Cluster Architecture Basics7/7
most read this week
posts (112)
- Part II: Standardizing AI in the SDLC Across the Organization10/10
What a standardized Agentic SDLC looks like
The Implementer Agent's final flow from ticket to PR, the shared layers the organization provides, what still belongs to the team, and a map of tools across…
The OSI Model in the Cluster and the Cloud: Who Sees What When You Open a Page
L3, L4 and L7 are shorthand for how deeply a piece of infrastructure looks into traffic. One request, from clicking "Pay" to the payments pod, layer by layer.
- Part I: From Prompts to Agentic SDLC9/9
When Managed Agents are not enough: a platform for running agents
The ninth level shows the moment when an agent stops being only a Claude session and becomes a platform workload with IAM, networking, secrets, observability…
- Part II: Standardizing AI in the SDLC Across the Organization9/10
Evals and quality gates
Observability tells you what the agent did, not whether it did it well. Evals next to the agent code work as regression tests for behavior and block rollout…
- Recovery from the Ground Up: Rollback, PITR, and Returning After Failure5/5
After Recovery: Green Dashboards Are Not the End
Recovery ends the impact of an incident, but it does not end the work. We still need to reconstruct the timeline, improve the runbook, and check whether the…
- Part II: Standardizing AI in the SDLC Across the Organization8/10
Observability, security and control
With dozens of agents you need every model call, tool call and RAG document as a span in a trace, control over input and output at the tool boundary, and…
- Testing Through Layers: From Code to Infrastructure9/9
Infrastructure Tests and Security Tests: Configuration Before Production
An infrastructure test checks environment and configuration, while a security test guards risks that are often invisible in application code.
- Part II: Standardizing AI in the SDLC Across the Organization7/10
Human-in-the-loop: when the agent cannot continue on its own
Sometimes the next decision needs information that is not in the repo, RAG or the ticket, or a decision the organization does not delegate to a model. The…
- Part II: Standardizing AI in the SDLC Across the Organization6/10
Runtime, state and memory
A real task gets interrupted by failing tests and approvals. The runtime says where the agent runs, state lets one task resume, and memory carries knowledge…
- Part II: Standardizing AI in the SDLC Across the Organization5/10
MCP, tools and central access
Instead of a local MCP with every developer's personal PAT, the organization exposes a shared tool layer with agent identity, policy and audit.