Software solutions do not start with code
I can move from idea to code faster with AI, but the real work still starts before that: asking better questions, spotting constraints, and building the right mental model.
Shaping with AI
Categories
2026
Switching from doing to directing
As AI takes over more of the mechanical work, engineers at every level need to learn how to direct, verify, and own what gets shipped. This article reflects on why that mindset should start early, not only at senior level.
Shaping with AI
How I used Cortex Code Desktop to assess the risks in Snowflake's upcoming behavior change bundle
Snowflake rolls out semantic and breaking changes through behavior change bundles. Before a bundle becomes the default, there is a window to test how those changes might affect your workloads. In this article, I show how I used Codex and later Cortex Code Desktop to assess the risks, identify likely breakpoints in our Airflow and dbt pipeline, and map out likely fixes before the bundle lands.
Shaping with AI
Effectiveness anti-patterns — part 6: YAGNI: don’t abstract before the second use case
A practical YAGNI story from data engineering: keeping a CI workflow intentionally narrow until the second real use case appeared.
Career notes
From principles to production: starting a new series
Why AI makes engineering fundamentals more important, not less, and why I am starting a series on the principles that still shape real production decisions.
Technical notes
How I improved AI-generated Python code by setting clear instructions
A reflection on turning one AI-generated Python prototype into reusable coding-agent guidance: clearer implementation boundaries, stronger quality expectations, and a shared quality bar before prototypes become real tools.
Shaping with AI
Effectiveness anti-patterns — part 5: when good work is invisible work
A reflection on invisible work in platform and data engineering teams: why success often looks like silence, why the work-speaks-for-itself instinct breaks at the senior-to-staff transition, and what I have tried to make my work more legible to people I don't sit next to.
Career notes
Parallel agents hit a human limit faster than a system limit
Running multiple coding agents in parallel feels like a speed win until the bottleneck becomes the person trying to supervise all of them.
Shaping with AI
Effectiveness anti-patterns — part 4: meeting overload
A reflection on meeting overload: why every invitation deserves real scrutiny, why three to four meetings a day make focused engineering work much harder, and why leaving the room when I cannot contribute can be the responsible move.
Career notes
How to set up a Claude Code agent team workflow: from parallel tasks to peer review
A practical walkthrough of setting up Claude Code agent teams with clear roles, task ownership, communication rules, review steps, and lightweight output contracts.
Shaping with AI
Orchestrating parallel subagents for data collection, exploration, and reporting
A hands-on case study of using Codex subagents to coordinate data collection, preparation, and Slidev report generation in parallel.
Shaping with AI
Creating agent-backed skills in Codex: a uv-first Python repo health check
I wanted Python repository health checks without flooding the main Codex thread with raw `uv`, `ruff`, `pytest`, and `ty` output. So I built a subagent-backed skill that inspects the repo, runs the usual commands, and returns a compact report with clear pass, warn, and fail signals.
Shaping with AI
Effectiveness anti-patterns — part 3: knowledge silos feel efficient until they don't
A reflection on knowledge silos in engineering teams: why they often start as responsibility rather than ego, and what helps spread context before one person becomes the queue.
Productivity
In agent systems, context matters more than the prompt
A practical guide to what the planner actually sees, why context is more than the prompt, and why better context selection matters more than simply adding more tokens.
Shaping with AI
Creating a Cursor rule to keep Python workflows on uv
I already had reusable `uv`-first instructions for Claude Code and Codex. I have now added the Cursor path too: generate a rule from the same source guidance so Cursor stays on the same Python workflow.
Shaping with AI
Effectiveness anti-patterns — part 2: technical debt denial
The second part of a series on engineering effectiveness anti-patterns, focused on what happens when we keep postponing system health in favor of short-term feature delivery.
Productivity
Agents in plain English for developers
A practical guide to what agents are, what they are not, and how to reason about them in terms of context, tools, host loops, and runtime constraints instead of hype.
Shaping with AI
Setting up a uv-first Python workflow for Claude Code and Codex
I kept noticing coding agents drifting into `pip install`, raw `python`, and hand-edited dependency files in Python projects. So I built a small repo with reusable `CLAUDE.md` and `AGENTS.md` guidelines that push both Claude and Codex toward a consistent `uv` workflow.
Shaping with AI
Effectiveness anti-patterns — part 1: context switching
The first part of a series on engineering effectiveness anti-patterns, focused on how constant context switching quietly erodes focus, depth, and overall effectiveness in day-to-day engineering work.
Productivity
Avoiding getting territorial pays off — in collaboration and trust
A reflection on territorial behavior in engineering teams: why it shows up, why it hurts collaboration, and how to catch it in yourself before it damages trust.
Career notes
The show must go on
The show must go on in the AI era. The responsibility - and the opportunity - for experienced software engineers to help grow the next generation is still here. Sharing knowledge and supporting early-career talent is something AI did not and could not change.
Community
PyLadies Budapest: Start writing your own functions in Python
A talk on understanding and using functions in Python — what they are, why they matter, and how to start writing your own — at PyLadies Budapest.
Community
Shifting from coding to building — What "coding is largely solved" could mean for engineers
Listening to Boris Cherny talk about “coding being largely solved” felt familiar — it describes a shift many of us are already living, even if we don’t always name it.
Shaping with AI
Claude Code now has built-in Git worktree support
Anthropic just shipped native git worktree support in Claude Code CLI. Last week I wrote about a hub-and-spoke worktree setup for parallel AI agents --- now it's getting baked right into the tool.
Shaping with AI
How I develop multiple features at once with AI agents and git worktrees
Git worktrees let you run multiple AI agents in parallel — each on its own branch, in its own directory, without stepping on each other. Here's how I use this when I've got a few features cooking at once.
Shaping with AI
When to use subagents and when to use agent teams in Claude Code
If you've been using Claude Code for a while, you already know **subagents**. They're the workhorses behind the `Task` tool — you send one off to research something, it comes back with results, and you move on. Simple, effective, and cheap on tokens.
Shaping with AI
The risk of climbing hard in the wrong direction
Career growth isn't just about moving faster or working harder. At some point, the real risk shifts: climbing efficiently in the wrong direction. As I think more seriously about moving beyond Senior, I've realized that focus — not ambition — is the constraint most of us underestimate.
Career notes
2025
The resources that kept me inspired and growing throughout 2025
As 2025 wraps up, I've been thinking about what actually helped me grow this year. Not everything—just the resources that stuck with me on my journey.
Career notes
The hidden causes of self-doubt at work
Over the last five years, as a Senior Data Engineer on the EM / Staff trajectory, and at times leading teams directly, I’ve learned that one of the most rewarding challenges isn’t just delivering results…
Career notes
A senior mindset is ultimately about trust
Many people still think that not knowing something is a weakness — that admitting it might make them look less competent. And in tech, this belief is even stronger.
Career notes
AI is a tool. Judgment is the job
Critical thinking is your most valuable asset in the age of AI-assisted development. Why judgment still matters when building with AI.
Shaping with AI
Clean, structured, multi-destination, non-blocking logging in Python
Still relying on print statements to debug your Python code? There’s a much better way: clean, structured, multi-destination, non-blocking logging.
Technical notes
PyLadies Budapest: Python data structures in practice
A talk on practical Python data structures — tuples, sets, and the collections module — at PyLadies Budapest.
Community