Asarel Alejandro Núñez Segoviano
Lead Software Engineer · Product Engineering
WEPORT (Radiant Logistics) · Mexico City · Spanish / English
human need ──▶ discovery ──▶ scope ──▶ contracts ──▶ review ──▶ measure
│ ▲
agents └─ build ────┘Ten years shipping software. Most of my attention now goes to making agent output something you can trust: I own the contracts and the review, agents work inside them, and a change isn't done until there's a number showing it moved something.
The work also goes past the codebase. IT infrastructure, security and governance, and helping teams outside engineering pick up AI and automation.
+ Owning initiatives end to end: discovery, MVP scope, build, deploy, adoption
+ Choosing the stack per project against technical and commercial constraints
+ Designing agentic workflows governed by TDD, OpenSpec and adversarial review
+ Building the harnessing layer that keeps agent context governed and reproducible
+ Tuning AI economics, measuring token cost against delivered capacity
+ Improving IT infrastructure alongside product: security, governance, centralization
+ Bringing other areas and stakeholders onto AI and automation
+ Driving adoption from C-level to operators, new hires and vendors2025 · now WEPORT / Radiant Logistics Lead Software Engineer · freight forwarding
2023 · 2025 VOXPOP Software Development Manager · retail streaming
2016 · 2022 VOXPOP Full-stack · AndroidTV, React Native, AWS
2018 · now Independent React / React Native consulting
Native modules, device fleets, streaming platforms, hardware, IT operations. Infrastructure
is still part of the job. Most of what I know about how systems fail, I learned there.Fundamentals are the part that carries over. Data structures, concurrency, protocols, how a system behaves under load. They hold across every language on this list, and they're what let me spot a wrong answer from an agent.
Memorizing a framework's surface stopped being the hard part. The hard part now is picking the stack that fits a problem on technical and commercial grounds, then getting an organization to move to it. What follows is what the work runs on today. I've gone deeper in some of it than in the rest.
Foundations Data structures, concurrency, protocols, system design, debugging from
first principles
Judgment Stack selection against technical and commercial constraints, MVP scoping,
trade-off calls under ambiguity, adoption and enablement
Organization IT infrastructure, security and governance, process centralization,
cross-area AI and automation adoption
Orchestration Claude Code, opencode, MCP servers, subagent routing, multi-agent
adversarial review, in-house harnessing layer (devsource)
Method TDD, Red-Green-Refactor, OpenSpec, spec-driven development, contracts first
Intelligence GitNexus cross-repo graph, CodeGraph per-repo index, Engram team memory
on a self-hosted sync server
Languages TypeScript, JavaScript, Node.js, Python, PHP, Bash
Interfaces React, Vite, Next.js, React Native, Tailwind
Services FastAPI, Flask, Pydantic, Express, Laravel
Data PostgreSQL, MySQL, DynamoDB, Athena, medallion data lakes
Cloud AWS Lambda / S3 / DynamoDB / Athena / IAM, Serverless, Docker, GitHub Actions
Quality Vitest, Jest, Playwright, DORA metrics, chaos testing, data-quality reporting
Domain Freight forwarding, retail streaming, internal platforms, integrations1. Prefer clear boundaries over clever abstractions.
2. Treat APIs and integrations as contracts.
3. An agent is only as reliable as its context. Govern the inputs, review the outputs.
4. Know the fundamentals well enough to tell when the agent is wrong.
5. Adopt on evidence, not on novelty. Measure the cost before committing to the tool.
6. Automate the boring parts, keep the important parts explicit.Generated daily from WakaTime and the GitHub API, so nothing here is typed in by hand. That also limits them to what those APIs can see. When a metric needs a caveat to be true, I leave it out.
agent_workflow — last 7 days
sessions 33 sessions · 292 prompts · 1.62k chars per prompt
agent_time 24h 15m (99.55% of tracked time)
lines_generated +20,054 / -147
context_moved 17.43M tokens in · 3.14M tokens out
leverage 826 lines per agent hour · 69 lines per prompt
context_cost 869 tokens in per generated line
top_surfaces Other 6h 25m · HTML 6h 10m · Markdown 3h 02m · PHP 2h 58m
delivery — last 30 days
prs_opened 170
prs_merged 168 (99% of opened)
pr_size median 362 lines per merged PR
lines_shipped +98,960 / -6,687
active_repos 10
contributions 515 (private included)
active_days 25 of 30 days
streak 13 consecutive days
commit_rhythm — last 30 days · America/Mexico_City
morning 06-12 59 commits 🟦🟦🟦🟦⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛ 18.2%
daytime 12-18 223 commits 🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦⬛⬛⬛⬛⬛⬛ 68.6%
evening 18-24 42 commits 🟦🟦🟦⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛ 12.9%
night 00-06 1 commit ⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛ 0.3%
pace — since 2019-05-24
tracked_total 2274h 16m across 7.3 years
last_30_days 54h 38m (12h 44m per week)Updated 2026-08-30 10:46 UTC
This Week I Spent My Time On
Programming Languages:
Other 6 hrs 25 mins 🟦🟦🟦🟦🟦🟦🟦⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛ 26.32 %
HTML 6 hrs 10 mins 🟦🟦🟦🟦🟦🟦⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛ 25.32 %
Markdown 3 hrs 2 mins 🟦🟦🟦⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛ 12.46 %
PHP 2 hrs 58 mins 🟦🟦🟦⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛ 12.22 %
SQL 1 hr 42 mins 🟦🟦⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛ 06.99 %
Editors:
Claude Code 23 hrs 47 mins 🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦⬛ 97.64 %
Opencode Cli 18 mins ⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛ 01.30 %
VS Code 15 mins ⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛⬛ 01.07 %
Operating System:
Linux 24 hrs 22 mins 🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦 100.00 %
Last Updated on 30/08/2026 10:46:40 UTC
The split between developer, product manager and QA is dissolving. The product engineer becomes the base unit: one person carries a business need from discovery through to a measured outcome, with agents doing the mechanical work inside contracts a human owns. Teams get smaller and their scope gets wider.
This changes the job itself. Getting a model to write code turned out to be the easy part. The work is in making that leverage trustworthy: context you govern, output you can audit, data the organization can decide with, and a number at the end that shows whether any of it mattered.
It moved what depth means, too. Fundamentals still decide outcomes, because they're how you catch a wrong answer. Framework trivia doesn't, because that's the part the agent covers. The scarce skills are reading a business need, picking a stack that fits it commercially as well as technically, and getting people to adopt the change. Language choice is an implementation detail now.
What I'm working toward is turning this into a discipline. Standards, guardrails, enablement, so the practice survives being handed to someone else. Right now it's mostly folklore, passed between people who happened to figure it out.
Two things I don't buy. Autonomy without governance: broad access with no contracts, acceptance criteria or audit trail gives you speed you can't trace. And metrics theater: I removed a lead-time metric from this page because the number was accurate and measured nothing.