Evidence

Fig. 01 · The ledger

Show the work.

Seven products plus this site: eight AI-orchestrated builds in flight, across five unrelated domains, so the model can't pass by accident. Real screens, real shipped features, real lessons.

Building
ForgeCareer Job Hunt: 24,623 live listings pulled from 416 sources
ForgeCareer
Consumer SaaS · Web · auth · SMS

A career lifecycle platform, where the résumé is just a feature.

Shipped
✓ OTP phone verification
✓ Versioned Terms consent
✓ ATS score explainer
✓ Job feed ingestion
What it tests
▸ Proves: the orchestration can run a regulated consumer SaaS end to end, with auth, SMS consent, and versioned Terms documented before the first user, where a vibecoder ships a demo and skips compliance
▸ Beating ATS scoring and AI-content detectors without gaming them
▸ Coaching that adapts to the person, not a template
Building
ForgeMoneyHawk: 472 receipts for March, categorized, with deductibility and tax savings
ForgeMoneyHawk
SMB fintech · Web · OCR · email ingestion

Receipt and expense intelligence for small operators: scan the inbox, read the receipt, categorize it, and know your tax position by month end.

Shipped
✓ Inbox scanner for emailed receipts
✓ OCR with a human review queue
✓ Deductibility rules per category
✓ Tax summary and export
What it tests
▸ Proves: the orchestration holds up on financial data, where every categorization is traceable and a human review queue is part of the design, not a patch after the first wrong number
▸ OCR accuracy against a review queue that stays humane
▸ Turning categories into a tax position, not a spreadsheet
Experimenting
ForgeWoWBot Bot Commander: build a bot, build a party, build a 25-bot raid, in-game
ForgeWoWBot
Game systems · Lua · server

Bot Commander for private MMORPG servers: build parties and 25-bot raids by role, class, and strategy, from inside the game.

Shipped
✓ Bot Commander: quick build, roster, targets, formation
✓ Party and raid composition templates
✓ Raid-wide command bar: follow, attack, retreat, regroup
What it tests
▸ Proves: the orchestration can build inside a hostile stack, Lua on a 20-year-old server core, with no docs, no framework, and no tutorial to copy
▸ Real-time analysis of a live world, not request/response
▸ Composition logic: roles, classes, strategies that hold up in a raid
Experimenting
ForgeWoWGM GM Toolkit: accounts, items, travel, quests, and server control from inside the game
ForgeWoWGM
Game systems · Lua · AzerothCore

Game Master toolkit for AzerothCore servers: every admin command behind a searchable, in-game UI.

Shipped
✓ GM Toolkit: accounts, players & items, travel, quests, server
✓ Teleport, summon, item sets by ID
✓ Command search across every GM command
What it tests
▸ Proves: the orchestration can map and wrap a very large existing system it doesn't control, every command inventoried, then ship a governed UI over it
▸ Admin power with guardrails: quick toggles, explicit targets
▸ Shipping inside someone else's client
Defining
Player
"Where did the Hungerer go?"
Private server
bot hears it · who, where, quest, time
Orchestrator
pulls context, in parallel
LLM
composes, in character
Background calls
lore · Thaelis
persona · Nokpep
world · night, raid on
player · lvl 9, quest 1/3
Bot, in character, about a second later
"Thaelis? Slunk off toward the Arcane Cores when the moon came up. Bring a healer. I'm not it."
✓ lore · quest state · time of day · player level · temperament
ForgeWoWRP
Game systems · AI in the product · Lua · server · LLM

A private server wired to LLMs so every bot can hold a real conversation: grounded in the game's lore, the world state, current conditions, and the human in front of it.

Shipped
Nothing yet. Defining the problem first.
What it tests
▸ Proves: the orchestration can put an LLM inside the product, not just behind the build, with latency, cost, and coherence budgeted up front
▸ Do we still need writers for every line when the lore already fills books and films?
▸ Keeping a thousand bots coherent with one world state
Prototyping
WAVE 07 / 20 ◆ 1,240 ♥ 18 ▶▶ 2×
SPAWN →
→ BASE
NEXT WAVE
12 enemies · 2 bosses
armor ▲ · speed ▲
in 0:14
◆120
◆200
◆350
3D prototype · screenshots soon
ForgeTD
Games · Unity · C#

A Mac tower defense game in Unity, aimed at Steam.

Shipped
✓ Procedural map generation
✓ Physics tower targeting
✓ Wave difficulty tuner
What it tests
▸ Proves: the orchestration can carry story, art, code, and deploy as one line in a stack the builder has never used, Unity and C#, and still hit a taste bar, not just a compile
▸ Does the loop feel good, not just work
▸ AI-generated levels that still feel authored
Defining
Weekly check-in
Pairing
✓ both consented
Important dates
Wireframe · screenshots soon
ForgeCouples
Sensitive personal data · Mobile

A private space for two: notes, moods, challenges, and quick check-ins that turn patterns into positive habits. Both people opt in.

Shipped
✓ Mutual-consent pairing
✓ Weekly check-in survey
✓ Important dates tracker
What it tests
▸ Proves: the orchestration starts from consent and privacy in the requirements, on the most sensitive data there is, where a vibecoder starts from the screen
▸ Relationship habits built on positive values, not blame
▸ Spotting patterns early, turning them into small acts
8 in flight, this site included · updated continuously
Fig. 02 · Field notes

Learn it while we learn it.

What held, what broke, what we'd tell you on Monday.

Operating model № 01

Why we brought waterfall's paperwork back into an AI workflow

SOPs, architecture decisions, and quality gates were dropped because they were slow. With AI they cost minutes, and they are what keep a fast loop from becoming vibe coding.
ForgeCareer № 02

Designing versioned legal acceptance as a product feature

Terms, privacy, OTP consent, and re-acceptance interstitials: how compliance became a first-class part of the build, not a footer.
MMORPG tooling № 03

Bots that tell stories: using AI for character, not code

What changes when the AI is inside the product instead of building it: limits, latency, and keeping a live world coherent.
Get the notes by email One note per real lesson. No filler.
Fig. 03 · Want to talk more?

Seen enough? Three ways to take the model home.

01 · Diagnostic 2 weeks

The read

Ten conversations across your C-suite, product, and engineering. Your cost math, in your numbers. One page on what to change first.

You walk away with

A decision: where the AI-orchestrated model pays off in your org, and where it does not yet.

Fixed fee Book the read
Most orgs start here
02 · Pilot 90 days

One product, this way

One real product. We stand up a Product Builder + Architect pair, the three AI layers, the SOPs and gates. Shipping weekly by week three.

You walk away with

A shipped product, the playbook that built it, and two people in your org who can run it again.

Fixed fee + AI spend at cost Start a pilot
03 · Lead it Ongoing

Head of Product who builds

Fractional or full-time. Portfolio strategy, the operating model rolled across programs, and we keep building alongside your teams.

You walk away with

A product organization that ships with AI and can prove it to the board.

Retainer or hire Start the conversation