Zachary Proser

A trust-boundary threat model for prompt injection in tool-using agents, with capability scoping, external authorization, and testable security metrics.

A coordinate-consistent visual guide to diffusion and flow: forward noise, learned denoising, guidance, latent compression, DiT, sampling, and flow matching.

A benchmark score is manufactured, and every step leaves a trace.

What is an attention head, and how would you prove one does what you think it does?

Every frontier model runs the same circuit.

From open-domain question answering to production RAG: ingestion, chunking, BM25 and dense retrieval, reranking, context assembly, citations, evaluation, and failure tracing.

From the distributional hypothesis and word2vec to contextual sentence encoders: how embeddings are trained, compared, indexed with HNSW, migrated, and evaluated in production.

Transcribe Zoom meetings without a bot joining the participant list.

Transcribe Microsoft Teams meetings without a bot, admin setup, or an extra participant.

Transcribe in-person meetings on iOS or Android.

Transcribe Google Meet calls without adding a bot to the participant grid.

Transcribe client calls without a visible meeting bot.

How I use Granola to reduce meeting anxiety, check what I missed in real time, keep track of decisions, and stay present in every call.

How I route Granola meeting notes and transcripts into AI agents that draft posts, extract action items, create tickets, and write follow-ups.

I'm partly hard of hearing.

I replaced my portrait-centered homepage hero with the Mind on Fire mark rendered as living pixels, surrounded by every essay I've published as clickable constellations.

My verdict on eve, Vercel's open-source agentic framework: an emphatic yes.

Nick Nisi and I taught a hands-on hour at the AI Engineer World's Fair 2026: voice coding, agentic loops, verification gates, and scheduled tasks — one repo, four moves, and a live board showing the room reclaim hours of weekly toil in real time.

swyx told us our workshops are the gold standard for AI Engineer content.

How I run a solo AI consulting business from Slack: a fleet of single-purpose agents with the Claude tag as the meta-agent above them — reaching every surface, opening PRs, and maintaining the other bots — all behind a human approval gate.

My rule for AI agents: anything reversible runs unattended, anything irreversible or outbound stops for a human.

If you generate anything with a model in the loop, you need a step that reads the finished draft against a list of AI tells and rewrites every hit.

I reviewed Vercel's eve agent framework by building a real three-agent business team for my site — content, ops, and growth bots that deploy to Vercel and collaborate over Slack and eve channels.

My personal site runs like a small business with three AI employees — content, ops, distribution — and me as editor-in-chief.

Before a draft reaches me, my blog bot hands it to two cold readers from different model families that score it against a rubric and force a revision loop.

Claude Code writes the diff.

The unit of useful agent work is a reviewable artifact — a PR, a draft, a diff — not a chat reply.

Agent capacity is cheap now and getting cheaper; my attention is the fixed, scarce resource.

I went from blacking out on stage in front of 4,000 engineers to sleeping fine before jet-lagged days of talks abroad.

A stranger online said AI made him suspicious of all writing — that he now interrogates every sentence asking whether a machine wrote it.

You have about ten seconds to wow your user and convince them you're already solving their problem.

A year ago at WorkOS nobody could get a single Devin AI session working.

I wired my new place into an all-Apple smart home: Aqara cameras inside and out, a full eero mesh plus an outdoor repeater, Apple TVs, HomePod minis, smart plugs, and Siri automations.

Applying systems thinking to my internal experience.

I closed the WorkOS Applied AI Showcase with three learnings from shipping internal AI tooling: interface beats stack, complete the loop, and the imagination gap.

Gasoline and electricity, the line that won't dim.

We called him Intensity.

What my AI inbox-screener actually does to the templated cold outreach hitting my inbox.

A human-in-the-loop AI inbox classifier that drafts replies, with a per-category graduation ladder that lets each earn its way to auto-send.

I shipped a partnerships page.

Detailed comparison of WisprFlow mobile app versus Voice In browser extension for voice-to-text productivity across different platforms and use cases.

How real estate developers use WisprFlow to dictate project reports, investor communications, and property analysis while managing complex development timelines.

How mortgage brokers use WisprFlow to dictate loan files, document borrower communications, and streamline mortgage processing with voice-driven efficiency.

How mobile developers use WisprFlow to write React Native, Swift, and Kotlin code by voice - debugging on device, API integration, and mobile-specific development workflows.

How immigration attorneys use WisprFlow to dictate case notes, client narratives, and legal briefs while managing high-volume caseloads and multilingual documentation needs.

How CPAs use WisprFlow to dictate tax returns, client consultations, and financial analysis while maintaining accuracy in complex accounting workflows.

How content marketing managers use WisprFlow to dictate blog posts, social media content, and marketing campaigns while managing editorial calendars and team coordination.

How auditors use WisprFlow to dictate findings, document controls testing, and prepare audit reports while maintaining accuracy in complex financial examinations.

Detailed comparison of Granola and Zoom AI Companion for meeting transcription, intelligence, and workflow integration - features, accuracy, and pricing analysis.

How UX researchers use Granola's AI to capture user interviews, analyze behavior patterns, and transform qualitative research with intelligent conversation analysis.