Zachary Proser

An agentic terminal, a Mermaid fine-tune, and my context engine ctx.

I built pt, an attention-first agentic terminal in Go: durable work contracts, parallel agent lanes with byte-verified delivery, workers on rented orbs.

A complete record of 60 coding-agent runs across an 80.76 GiB local DeepSeek quant, hosted DeepSeek, and hosted Claude: task design, hidden scorers, repair passes, provider recovery, exact cost, uncertainty, and workload fit.

The strongest predictor in a 60-run local-vs-cloud coding-agent benchmark wasn't model quality or price — it was patch footprint.

A measured comparison of OpenRouter and Vercel AI Gateway across fees, provider redundancy, model catalogs, prompt caching, routing controls, privacy, and developer experience.

A current Wispr Flow vs Speechify comparison for dictation, reading, pricing, platforms, privacy, and daily professional writing.

Compare Apple Dictation with Wispr Flow for iPhone voice typing, including offline use, AI cleanup, personalization, privacy, and cost.

Two weeks using Vercel AI Gateway for Claude Code: measured cost and latency, plus the model table and alias that make switching models fast.

The best Granola alternatives for saved recordings, CRM automation, Notion workspaces, video review, and Zoom-native governance.

Sampling Thinking Machines’ Inkling through Baseten and Vercel, then using one alias pattern to route Claude Code between private local DeepSeek and the Vercel AI Gateway.

Eleven agents on Claude Opus 4.8 were burning more than $150 a week, so I shut them off.

The short version of two weeks spent measuring local and cloud inference: what a 128 GB MacBook is actually good for, why Vercel's AI Gateway is currently winning my DeepSeek workloads over DeepSeek's own API, which work goes to which model, and the one p…

Set up Granola with Zoom, understand its local capture and cloud transcription path, obtain consent, and troubleshoot missing audio.

Compare Granola's editable meeting notes with Zoom captions, retained transcripts, recordings, and AI Companion workflows.

Compare Granola's cross-platform private notes with native AI meeting notes in Microsoft Teams and Google Meet.

Compare Granola and Read AI across bot-free capture, meeting reports, recordings, speaker labels, coaching, consent, pricing, and the workflow each serves best.

A current Granola vs Otter comparison covering bot-free capture, live transcripts, saved recordings, collaboration, and the note-first workflow I prefer.

A current Granola vs Notta comparison covering bot-free capture, meeting bots, editable notes, recordings, languages, consent, pricing, and who each fits.

Compare Granola's dedicated meeting notebook with Notion AI Meeting Notes for capture, organization, consent, and sharing.

A current Granola vs Notion AI Meeting Notes comparison: capture, editable notes, organization, consent, sharing, and which workflow fits.

Compare Granola's cross-platform private notes with Microsoft Copilot and Teams Recap, including capture, governance, and workflow tradeoffs.

A current Granola vs Krisp comparison across bot-free notes, live audio cleanup, recordings, cloud processing, and real meeting workflows.

Compare Granola's private meeting notebook with Fireflies' bots, device capture, shared transcripts, uploads, analytics, and team workflows.

Compare Granola's private meeting notes with Fathom's bot and newer bot-free capture, video, clips, and team call workflows.

Granola is not currently HIPAA compliant.

A practical guide to Granola integrations: meeting-platform support, the Chrome extension, speaker tags, Notion, Slack, Zapier, CRM, MCP, and API limits.

A practical three-month Granola evaluation plan for new users who start through my affiliate link.

A documented review of Granola's cloud data flow, SOC 2 status, storage, SSO, retention controls, and current HIPAA limitation.

A practical guide to consent, disclosure, retention, and data handling when using Granola to transcribe business meetings.

My daily-use Granola review: how capture and cloud transcription work, what the note-first workflow gets right, and where another tool fits better.

Compare Firecrawl's managed web-data API with Crawl4AI's self-hosted Python crawler across setup, control, scaling, data handling, and cost.

A generic blueprint for designing AI evaluations: define the decision, version the system, build representative task sets, choose valid scorers, measure uncertainty, analyze failures, and connect offline tests to production evidence.

Compare Google Recorder, Samsung Voice Recorder, and Wispr Flow for audio recording, transcription, and system-wide Android dictation.

Choose the best AI meeting-notes app for editable notes, shared transcripts, CRM workflows, Notion projects, or video-call review.

After dictating 182,718 words across 36 apps, here is my honest Wispr Flow review: speed, accuracy, pricing, platform support, and drawbacks.

Wispr Flow costs $15 monthly or $12 with annual billing.

What does it actually cost to run a frontier-class open model?

How Granola captures meetings without a visible bot, turns transcripts and rough notes into structured notes, handles privacy, and fits real workflows.

Choose Firecrawl or Playwright for web data, scraping, browser interaction, and AI pipelines.

Compare Firecrawl and Apify for AI data: clean output, browser control, Actors, maintenance, pricing models, and the workloads each fits best.

Compare Firecrawl, Playwright, Puppeteer, Beautiful Soup, and Cheerio for AI web data: output quality, control, infrastructure, and maintenance.

I compared Granola, Otter, Fireflies, Fathom, and Zoom for bot-free notes, live captions, CRM workflows, price, and the meetings each fits best.

DeepSeek's retrained V4 Flash 0731 moved the model pricing baseline for everyone.

Open-weight speech synthesis got good enough to replace a paid API.

A model's name is a datasheet most people never learn to read.

Every agent I use — terminal, browser, phone, across Anthropic and OpenAI — started every conversation knowing nothing about me.

Hate mail from the edge of applied AI.

A complete working drawing for an AI training workshop: how to design, build, deliver, and follow through on an AI workshop for engineers or corporate AI training that works.

How tokenization works: BPE tokenizer training, ordered merges, byte fallback, token IDs, costs, multilingual tradeoffs, and why LLMs use tokens.

A worked guide to LLM inference memory, KV-cache paging, continuous batching, prefill, decode, quantization, and latency-capacity planning.