nikema/works in progress

I build things.
Then ask better
questions.

Case studies in building, testing,
and figuring things out.

Build, test, learnAn overlapping cycle of building, testing, and learning. Small experiments and feedback lead to better questions.BUILDTESTLEARNIDEAS →EXPERIMENTS→ BETTER QUESTIONSSMALL EXPERIMENTS.REAL FEEDBACK.PEOPLECONTEXTTRADEOFFSthe interesting partis the why

01 / Selected work

A closer look at the process.

MotionBrief

The five-second video I stopped generating

I set out to generate vertical video with AI. The footage was slow, expensive, and never matched the motion I wanted — so I generate the still and move the camera instead.

Open the field notes
MotionBrief's Create the visual screen for the Field Notes campaign, showing the visual direction and image prompts beside the generated golden-hour desk still, with Slow push in selected.
The visual step in Field Notes: Building in Public. A generated still and a selected camera move share one screen.

Hermes Hub

The office my agents built for themselves

I run a team of six AI agents. Their work was scattered across chat sessions and folders, so I assigned two of them to design and build a private dashboard to catalog it all. I directed the work, reviewed the merges, and tested the builds.

Open the field notes
Hermes Hub dashboard showing project stat cards, Continue Building, Recently Added, Quick Launch, and Needs Attention sections.
The deployed Hermes Hub dashboard with two active projects.

Social Content Agent

An AI workflow before an AI agent

I set out to build an agent that could create and post social content. My first version became an AI-assisted workflow.

Open the field notes
Social Content Agent draft library showing topic cards and filters for draft, approved, scheduled, and simulated published states.
The draft library, from an earlier version of the app. Published items reflect the simulation; no posts were sent to a social network.

Jev Job Screening Evaluation

I built the test before the automation

I got early access to Jev, TypeSafe's System One decision model. I tested it as a job-posting screener: a blind evaluation against my career copilot across 65 postings, sealed verdicts, and an evidence review of every disagreement. It was safe, but not ready.

Open the field notes
Cover of the Jev Phase A final report. Headline "Jev Phase A: Safe, but Not Promotable", the verdict that evidence was insufficient for promotion, and four metrics: 65 unique postings adjudicated, 0 false negatives, 13 of 13 rejects confirmed correct, 13 of 20 natural rejects required.
The Phase A final report cover: the verdict and the numbers at a glance.

CBS Fact Post Generator

Exa gives you evidence, not verdicts

A single-user tool that turns California Black history topics into verified, export-ready Facebook posts. OpenAI proposes claims, Exa retrieves sources, and a separate audit step judges each part of a claim against the excerpts before anything is composed.

Open the field notes
The fact post generator app showing a verified claim about Black Gold Rush miners at Negro Bar, with the generated caption, five hashtags, the live 1080x1080 image preview, and saved history.
The app in use: a verified fact about Black miners at Negro Bar, composed with caption, hashtags, and the exported image preview.

The decisions behind
the interface.

01

Start with a question.

What does the person on the other side of the screen need?

02

Follow the friction.

Look closely at the places where a good idea becomes difficult to use.

03

Show the reasoning.

Make room for the tradeoffs, the rough edges, and what comes next.

Outside the case studies

More thinking, in public.

Notes on the things I’m learning and making.

Read on nikema.dev ↗Same curiosity,
different pages.