Anthony AI Builds
antderosa@gmail.com

Useful AI systems, built from lived editorial problems.

25 years building editorial operations — now building the tools that run on top of them.
5AI-assisted builds in active use
25yEditorial and newsroom operations
0Slideware-only concepts
1Builder, user, and product owner

I spent my career running newsrooms and editorial teams — Reuters, The Wall Street Journal, The Daily Show, and most recently as VP of Newsletters at Pluralsight. Over the past several months I've been using Claude Code and Codex to build the systems I always wished existed: tools that turn scattered personal and professional information into something a person — or an editorial team — can actually act on.

None of this is theoretical AI-tool dabbling. Each project below was built end-to-end and is in active use. The demos below run on fictional data so you can click through them without me handing over my actual contacts, health logs, or calendar.

Start with an annoying real workflow

Every build began as a thing I was already doing manually: prepping calls, tracking leads, reading news updates, or interpreting health logs.

Let AI handle synthesis, not ownership

The useful part is judgment at the edges: turning messy inputs into ranked next steps, briefings, linked pages, and coaching notes.

Ship the smallest working system

These are personal products with real interfaces, data paths, and habits around them — built to be used, then improved.

Selected builds

Each demo uses synthetic data, but the product behavior mirrors the real tools.

01Context layer

AnthonyWiki

Claude CodePersonal knowledge baseSkill / agent system

A personal knowledge wiki compiled from journals, notes, and messages — not a filing system, but a writer's pass over raw personal data that synthesizes understanding into structured articles. It's the context layer that powers the two tools below: when Chief of Staff or Networker need to reason about who I am, what I care about, or who I know, they read from this wiki.

What it proves: AI is most useful when it can inherit durable context instead of starting every session from zero.
Synthetic data — fictional person, company, and project
Project: Lighthouse Onboarding Redesign
projectperson:Maya Okaforcompany:Brightline Learning

Maya Okafor proposed cutting the new-user onboarding flow from 9 steps to 4 after the March churn review showed 38% drop-off before first activation. She's running the redesign with the growth pod through Q3, reporting into Devon Price.

Anthony connected Maya with a former WSJ colleague, Priya Shah, now doing UX research at a competitor — Maya wanted outside perspective on activation metrics before presenting to leadership.

02Ops layer

Chief of Staff

Claude CodeDaily opsBriefings

A working daily-briefing and project-tracking system built on top of AnthonyWiki. Generates morning briefings, tracks open items across projects, preps 1:1s, and maintains a running memory of people, projects, and commitments — the kind of operational layer a chief of staff would maintain, automated.

What it proves: a personal AI system can become operational infrastructure when it remembers commitments and turns them into the next useful action.
Synthetic data — fictional team and projects
Today's Briefing — Wed, Jun 24
  • 1:1 with Devon Price at 10am — come with a take on the onboarding redesign timeline.
  • Lighthouse launch retro is overdue by 3 days — Maya is waiting on your notes.
  • Vendor contract renewal (Heron Analytics) needs a decision by Friday.
Open ItemOwnerDue
Finalize onboarding A/B test planMaya OkaforFri Jun 26
Send Heron Analytics renewal decisionAnthonyFri Jun 26
Q3 roadmap review with leadershipDevon PriceMon Jun 29
03Relationship intelligence

Networker

Claude CodeDecision supportJob search

A tool for deciding who in my network — and who in my network's network — is worth reconnecting with. Built it while running my own job search: it researches contacts and companies, tracks meeting notes and leads, and helps me prep for calls with actual context instead of generic LinkedIn skimming.

What it proves: AI can turn a fuzzy network into a ranked operating system for outreach, follow-up, and second-degree opportunity discovery.
Synthetic data — fictional contacts, no real names
Your network — ranked to reconnect
NameRoleTierLast touchWhy reconnect
Priya ShahUX Research Lead, Heron AnalyticsHot2 mo agoActive hiring manager, adjacent industry
Marcus LinVP Content, Fenwick MediaWarm9 mo agoFormer teammate, growing team this quarter
Dana OseiEditorial Director, Stratus GroupCold2+ yrs agoStrong alumni tie, no recent signal
Second-degree — people they know, you don't
NameRoleConnectorWhy intro
Leila HartmanHead of Talent, Heron Analyticsvia Priya ShahOwns hiring for roles matching your background
Tomas ReyesFounder, Northbeam Mediavia Marcus LinEarly-stage, actively building editorial team
04Feedback loop

HealthDash

CodexPokeNotionDashboard

I text what I eat and what exercise I do (running, strength training) to Poke, which feeds the data into Notion. HealthDash pulls that into a web dashboard with running coach-style commentary — AI giving feedback on the data, not just visualizing it.

What it proves: lightweight capture plus interpretation beats a dashboard that only shows numbers.
Synthetic data — not Anthony's actual logs
2,140
Avg cal / day
5
Workouts this wk
118g
Avg protein / day
Mon
Tue
Wed
Thu
Fri
Sat
Sun
Ran 4.2 mi, 38:12 — easy pace7:14am
Breakfast: oatmeal, banana, coffee — ~420 cal7:52am
Strength: legs, 45 min6:30pm
Coach note: Protein's trended below target three days running. Add a shake or extra serving of protein post-run tomorrow — your strength sessions will hold up better.
05News product

Chronicle News

CodexNews productAtomic content

A modern take on Circa — the mobile news app I helped build years ago that broke stories into discrete, structured "atoms" of information readers could follow over time instead of re-reading a full article on every update. Chronicle News rebuilds that idea for today: atoms of information assembled to get readers oriented on a story fast. Two sides of the same product — the editorial CMS where atoms get written and published, and the reader app where they get followed.

What it proves: editorial product instincts pair naturally with AI-assisted structured content and reader-first information design.
Synthetic data — fictional news story

Regional transit authority approves fare overhaul

2:41pmBoard votes 6–1 to approve the new zone-based fare structure, effective Sept 1.
1:15pmPublic comment period closes; advocacy groups split on flat-rate vs. zone pricing.
TueBackground: current flat-fare system unchanged since 2014, doesn't reflect ridership patterns.

Mid-size retailer reports surprise Q2 beat

9:32amShares up 11% pre-market after earnings beat consensus by 14 cents/share.
YesterdayAnalysts had flagged inventory overhang as the key risk heading into the print.