Selected work.

Things I've built or run that aren't slideware. Most live somewhere between "shipped" and "still hardening." Where I can share specifics, I do; where I can't, the framework's still public.

maven-template — AI Chief of Staff for Claude Code — project screenshot

maven-template — AI Chief of Staff for Claude Code

shipping
2026 · role: Architect & designer · audience: Claude Code power users

Public, open-source version of MAVEN, the internal AI assistant I built and deployed across MMA's 90-person global org in 16 countries. An opinionated Claude Code template that turns Claude into an AI chief of staff for any organization: templated org context filled via interactive /onboard, session continuity across days, custom statusline with cost/pace tracking, 5 multi-agent workflow patterns, and curated MCP recommendations. Goes beyond Sterling Chin's MARVIN template with org-awareness and power-ups.

Claude Codetemplated org context/onboard commandcustom statusline
  • Public on GitHub — open source, MIT licensed
  • Templated org-context.md + org-brand.md with /onboard walkthrough
  • ccusage-calibrated statusline with session + weekly cost/pace + reset countdowns
github.com/alectivism/maven-template ↗
Cross-Model Agent Delegation — project screenshot

Cross-Model Agent Delegation

shipping
2026 · role: Author & maintainer · audience: Claude Code power users

Orchestrator-worker kit for Claude Code that cuts expensive-model token burn: route research, review, retrieval, and mechanical edits to cheaper Sonnet and Haiku subagents, plus GPT-5.6 through the Codex CLI, while orchestration stays on the top model. Six task classes with pinned models and reasoning effort, agent templates that stop ad-hoc subagents from silently inheriting an expensive session model, and an optional hook that stops raw Codex CLI calls from skipping the wrapper.

Claude CodeCodex CLIskillsagent templates
  • Extends Anthropic's Fable→Sonnet delegation benchmark (96% of the accuracy at 46% of the cost on BrowseComp) with a second GPT-5.6 worker pool
  • Spreads load across separate Claude and ChatGPT/Codex quotas, so your Claude limits last longer
  • Six deterministic routing classes; MIT-licensed, installs by copying one skill into ~/.claude/skills
github.com/alectivism/cross-model-agent-delegation ↗
Marketing Truths — evidence platform + chatbot — project screenshot

Marketing Truths — evidence platform + chatbot

production
2025–26 · role: Architect & lead engineer · audience: marketing leaders + strategists

Evidence-first platform and AI chatbot that answers marketing strategy questions from peer-reviewed science, with sources attached. 200+ empirical generalizations distilled from 10,000+ peer-reviewed studies across 90 marketing capabilities. Co-built with MMA and the Marketing Science Institute (MSI), live at truthsalpha.com.

Python 3.12LangChainChromaDBVercelSentry
  • Public platform + AI chatbot live at truthsalpha.com
  • 200+ empirical generalizations from 10,000+ peer-reviewed studies, 90 marketing capabilities
  • Co-created by MMA and the Marketing Science Institute; router-based RAG, Sentry-monitored
truthsalpha.com/ ↗
MMA Claude Skills bundle — project screenshot

MMA Claude Skills bundle

shipping
2025–26 · role: Designer, author, packager · audience: marketing & ops teams

Open-source pack of 21 Claude Skills tailored to marketing operations: asana-task, briefing-prep, case-study, content-draft, content-strategy, email-draft, event-promo, lab-summary, launch-strategy, meeting-followup, member-comms, press-release, research-brief, slack-summary, sharepoint-find, and more. Each skill is a self-contained markdown contract.

Claude Skills (MD + zip)Anthropic Skill SDK
  • 21 skills, ~3k LOC, deployed internally across MMA team
  • Used as reference for member orgs starting their own skill libraries
  • Plugin-installable via the MMA Plugin
/frameworks ↗
AI Leadership Think Tank (ALTT) — project screenshot

AI Leadership Think Tank (ALTT)

active
2024–26 · role: Co-program director · audience: senior AI + marketing leaders

300+ marketing and AI leaders from 180+ companies including GM, AT&T, Mastercard, Kroger, Home Depot, Coca-Cola, Unilever, and Hilton, whose CMOs sit on MMA's boards. I co-lead the community, and run the Future Labs experiment program and the Responsible AI Innovation Lab (RAIL). Continuous working sessions (not cohort-based).

communityFuture LabsRAILworking sessions
  • 29 controlled experiments across 18 global brands
  • Average +160% lift in advertising performance over baseline
  • 8 published Responsible AI guides adopted by Fortune 500 marketers
mmaglobal.com/marketing-arenas/ai ↗
AI Agent Foundry — project screenshot

AI Agent Foundry

active
2025–26 · role: Designer & lead · audience: MMA staff

Mandatory workforce transformation program at MMA: every employee dedicates 10% of capacity to building AI agents and automations. Tiered certification (Foundations → Agent Development → Agent Management). The output: 40+ live workflows across marketing, sales, CX, research, operations.

workforce programClaude Coden8nZapier
  • 100% of 90 staff trained on AI fluency basics
  • 40+ production workflows built by non-engineers
  • Workflow library: meeting transcripts → CRM updates, lead discovery, weekly reporting, branded content drafts
mmaglobal.com/ai-training ↗
Responsible AI Toolkit — project screenshot

Responsible AI Toolkit

drafting
2026 · role: Author · audience: engineering & RAI teams

Drop-in starter pack of bias evaluation rubrics, trace exporters (Claude Code → OpenTelemetry), PII filter hooks, and audit log schemas. For teams shipping multi-agent systems who need governance scaffolding without inventing it.

MD rubricsPython eval helpersMCP exporter
  • Open-source repo, ships with eight RAIL-published frameworks
  • Used as reference for two enterprise governance builds in pilot
github.com/alectivism ↗
Decoding AI for Marketers — project screenshot

Decoding AI for Marketers

shipping
2024–26 · role: Content support + intros · audience: marketing leadership + ops

MMA's flagship AI marketing training, designed and led by Greg Stuart and Rex Briggs. I'm not the program lead — I support the content and handle introductions. Worth highlighting here because it's where most ALTT and MMA AI work ladders up from: 6,000+ marketers trained, 2,000+ certified, plus a paired podcast running every ~3 weeks.

MMA flagship trainingpodcastAI certification
  • 6,000+ marketers trained, 2,000+ certified
  • Podcast with new episodes every ~3 weeks
  • Two cohorts running in 2026
mmaglobal.com/ai-training ↗
MMA Design System — project screenshot

MMA Design System

shipped
2025 · role: Designer & lead · audience: internal design + member orgs

Designed MMA's slide deck design system so staff can scale their design practices and consistently follow brand guidelines when using ChatGPT or Claude to create PowerPoints, documents, and visual assets. The interesting bit is the operational discipline — making brand consistency the default when AI tools generate visuals.

Slidesdesign tokensbrand framework
  • Adopted as MMA's default deck template for member-facing pitches
  • Brand-rule enforcement baked into AI workflows
ArcPush — project screenshot

ArcPush

active
2025–26 · role: Solo build · audience: self

Personal Pushbullet replacement. Routes URLs and notifications across my devices with Arc browser space targeting — a work link opens in my Work space, a personal link opens in my Personal space. Has a built-in RSS reader (auto-tagging by domain/path) that feeds my content pipeline. Single-user system, locked to one Firebase UID.

Chrome MV3ReactFirebase FunctionsReact Native
  • Daily driver on Arc browser + Android
  • RSS layer integrated with the content pipeline
  • Reference build for MV3 + Firebase + React Native
Content Pipeline — project screenshot

Content Pipeline

shipping
2025–26 · role: Architect & builder · audience: self + staff

Idea-discovery + drafting pipeline. A curated set of podcast RSS feeds is transcribed twice a week and Claude extracts *topic ideas* (not finished posts) into a Notion intake. I triage, then draft each post in my voice with optional research at draft time. The pipeline reduces the friction of "what should I write about this week" — the writing remains mine. The publish half is open-source via n8n-workflows.

Claude Coden8nNotionPythonLinkedIn APIX APIBluesky
  • Idea backlog stays full; topics emerge from real industry discussion
  • Multi-channel publishing (LinkedIn, X, Bluesky, Threads) shipped as importable n8n template
  • Structured to scale to MMA staff seats with separate channel configs
github.com/alectivism/n8n-workflows ↗
Board LinkedIn Monitor — project screenshot

Board LinkedIn Monitor

shipping
2026 · role: Architect & builder · audience: MMA CEO + regional CEOs

Monitors LinkedIn posts from MMA's 234-person board roster across 6 regional boards. When a board member posts, generates 2-3 draft comments that Greg Stuart (CEO) or a regional CEO could leave on the post — drafts queue to Airtable for human review. No auto-posting. Built around a strict cost ceiling: ~$0.25/week Apify spend with kill-switch + per-region enable flags.

n8nApify (LinkedIn scraping)SalesforceAirtableClaude
  • Full roster pipeline: 234 board members enriched with LinkedIn URLs via Salesforce + web fallback
  • n8n weekly cron + webhook for ad-hoc reviews
  • Cost-tiered rollout: global board only at $0.25/wk, scales to all regions at ~$0.55/wk
  • Slash command /board-linkedin-review for triaging draft comments
This site (alecfoster.com) — project screenshot

This site (alecfoster.com)

shipping
2026 · role: Architect, designer, builder · audience: speakers · recruiters · readers

Personal site. Designed to showcase agent-strategy work, host The Eval newsletter mirror, and serve as a reference for builder-tone marketing. Includes a live AI chatbot, a mirrored Beehiiv archive, and active analytics.

Astro 6Preact (islands)Tailwind 4MDXVercel
  • 16+ pages, 5 content collections, mirrored newsletter archive
  • Source open at github.com/alectivism/personal-website
  • Live Claude-Haiku chatbot scoped to my work
github.com/alectivism/personal-website ↗

Some of this work I can't share publicly without NDAs lifted. If you're evaluating me for an engagement and want the closed-door version with metrics, references, and architecture diagrams, say so.