Our Work

What we do. Research that goes deep and writing that gets to the point. We read the primary sources, cross-check every claim with more than one AI, and a human editor makes the final call. Then we cut until only what matters is left.

Example: the AI frontier report. What a frontier model is, who’s leading, and which model is worth paying for. Built from every major AI rating service, fact-checked by a second AI, edited by a human, and kept current every day.

What you get. Reports, articles, briefs and web copy, sourced and verified, in plain language for your readers and full technical depth for those who want it.

Read the frontier report →

Technical detail

• Multi-model research and cross-examination: one AI researches, another evaluates and fact-checks, a human edits.
• Primary sources first; every figure traceable to its source and date.
• A standing daily research agent that monitors sources and grades each finding before it’s used.
• Two-layer writing: a plain-language lead, technical detail beneath it.

What it is. A haunted house game built in Unreal Editor for Fortnite, for the millions of players on Fortnite. Portraits whose eyes follow you down the hall. A doorman who greets you at the gate. Scares that react to what the player does, and choices that change the night.

How we built it. One human set the creative direction. A team of AI teammates wrote the game code, built and placed the 3D characters, tested, and checked each other’s work, turn by turn. Work that usually takes a studio, done by a team of five seats.

Status. In production now. Build clips are coming to this page as the house comes together.

Technical detail

• Unreal Editor for Fortnite (UEFN), with gameplay logic written in Verse.
• Custom 3D characters generated, rigged and animated for the game.
• Player-tracking effects, event-driven scares and branching player choices.
• Automated build checks and a shared review process across the team.

The problem. Domains spread across three registrars, years of renewals on autopilot, and a new studio name that needed a home.

What we did, in one morning.

• Registered two new domains and had them live within minutes: one redirecting to the studio site, one to the AI Race Tracker, with email forwarding on the studio address.
• Moved three established domains onto one lean, modern registrar, approved for same-day completion instead of the usual week.
• Started a client’s domain on the same move, well ahead of its renewal date.
• Kept every site and every mailbox running throughout. No DNS changes, no downtime.
• Scheduled the last holdout’s move for its renewal window, so nothing ever lapses.

The result. Seven domains under control, auto-renew and privacy on, and one less vendor to pay.

Technical detail

• Registry lookups (RDAP) and DNS resolution checks to inventory registrar, expiry, nameservers and mail routing for every domain.
• Transfers sequenced to keep authoritative DNS on the existing hosting nameservers, so resolution never changed.
• 301 URL forwarding with tracking parameters; wildcard forwarding; mail forwarding on the new domain.
• Outbound transfers approved at the losing registrar to skip the five-day hold.

The problem. We inherited a web server running five live websites, ours and our clients’. The man who built it had passed away and left no handoff notes. Then a critical security flaw was announced in the language those sites run on, and four of the five were on a version that would never be fixed.

What we did, in one working day.

1. Snapshotted the server and backed up every database before touching anything.
2. Mapped everything we’d inherited: which site lived where and what it ran on.
3. Moved all five sites to a current, supported version, one at a time, each verified live before the next.
4. Brought WordPress, themes and plugins current, including a major upgrade on three sites.
5. Ran a deep security check against official published copies of every file. Clean.
6. Fixed what it turned up: a contact form that had never delivered mail now delivers end to end.

The result. Five sites patched, current and verified clean, and a server we now run and document ourselves.

Technical detail

• Droplet snapshot and per-site database dumps first; snapshot kept as evidence.
• Read-only inventory script, reviewed and hash-checked before its single run.
• PHP runtime migration off an end-of-life branch, lowest-risk site first, each verified live.
• File integrity checked against official checksums for core and plugins.
• Mail delivery fixed at the domain-authentication level; spam protection restored.

What it is. A live chart of the AI race. The leading rating services each score AI models their own way, so nobody can read them side by side. We built a single scale that puts every frontier model on one line: how many months behind the leader it is. One glance tells you who’s ahead, who’s closing, and which model is worth your $20 a month.

How it runs. An AI agent we built checks the rating services every morning. When the frontier changes, it updates the chart and alerts subscribers. Every result is logged, so the race has a history: weeks at number one, weeks in the top three, best rank.

Why it matters. AI moves weekly. A ranking from last month is already out of date. This one isn’t.

See the race →  ·  Shareable at airacetracker.com

Technical detail

• Normalization of independent leaderboards onto a common months-behind-the-frontier scale.
• Scheduled daily agent: collection, change detection, grading, and alerting on frontier changes only.
• Historical store with rank statistics computed by formula; charts rendered live from the data.
• WordPress delivery with a custom dark front end, deep links into the page, and email signup.