Services · Tech
AI levelled the
playing field.
Most businesses haven't noticed.
For a few hundred dollars a month in tool spend, you can now run functions that used to cost six figures to staff. Analyst-grade research. Designer-grade imagery. Copywriter-grade content. Dev-intern build help. Deep familiarity with the algorithms that became "AI" tells us one thing very clearly: the ones who notice this fastest are going to look unrecognisable on the other side.
The cost curve just collapsed
An unfair advantage hiding in plain sight.
This is the cheapest competitive edge of the decade. Most businesses are still treating AI as a content toy or a hype-cycle distraction. The ones who treat it as infrastructure are pulling ahead, quietly, and faster than the rest realise.
The lineage
Everyone started calling this AI recently.
We've been doing it the whole time.
The technology underneath what's now branded "AI" is the same machine learning, the same algorithmic automation, the same big-data thinking we've been working with since long before anyone made a Netflix documentary about it.
The new bit is that the tooling finally caught up to the ideas, and the API economics dropped to the point where anyone willing to learn the tools can use them. We didn't pivot into AI. We were already here. We were just waiting for the cost curve to make the obvious applications commercially sensible.
That waiting is over. The next five years belong to the businesses that act like it.
What we've already shipped
Six categories. All running in production.
Half of these power services we're already selling. The other half are the kinds of build we take on when a brief comes in.
Content engines
The content engine behind our SEO work, researchers, writers, editors, fact-checkers, QA and image producers each working with a custom AI co-pilot tuned to their specialty. The reason a small team can run across hundreds of sites at consistent quality.
See it in action →The Hutch
Editorial distribution at a cadence and quality the original agency model could never have justified. Per-site voice, per-site research, per-site imagery, all of it specified, trained and run as a system.
See it in action →Image & video pipelines
Production-grade asset generation for catalogue, lifestyle and editorial use. Trained against each brand's aesthetic. Built into the production stack so the same engagement covers the shoot, the AI-augmented variants, and the channel-ready deliverables.
Internal copilots
Retrieval over your CRM, knowledge base, product data, comms history. The version of 'AI strategy' that actually shows up at someone's desk every morning instead of living in a deck.
Agentic workflows
Repetitive operations work, research, classification, outreach prep, routing, triage, handed to an agent that runs against an evaluation harness so you can trust the output without re-reading every row.
Eval & governance
The unglamorous bit nobody else builds. Cost monitoring, drift detection, hallucination budgets, prompt versioning, output evaluation. The infrastructure that decides whether the AI work survives contact with a real production environment.
The traps
Four things that kill most AI projects.
We've seen every one of these in the wild. None of them are intrinsic to AI; all of them are intrinsic to how AI tends to get bought and built. Avoid the four and you'll already be ahead of most.
The demo isn't the product.
The hard part of AI work isn't the model. It's the integration into a workflow that real humans use under real pressure, and the eval harness that tells you whether it's still working in week six.
'AI strategy' is mostly a deck.
Most AI engagements end at recommendations. We treat that as halfway. The build, the deploy, the eval, the cost governance, the ongoing tuning, that's the work that earns the line item.
Output quality varies wildly per use case.
Some tasks are perfectly suited to LLMs and ship at human-grade quality on day one. Others look promising in a demo and quietly fail in production. Knowing which is which saves quarters of wasted effort.
Costs spiral if you don't watch them.
Token spend, model selection, caching, batching, prompt engineering for cost not just quality. Most AI bills double on the second month because nobody owns the meter.
The other half
Where humans stay.
AI is most useful when you're clear-eyed about where it doesn't belong. We're not building the version of YLO where the founders go on holiday and the bots run the company.
- Strategic tasteWhat's worth doing in the first place. No model has that opinion.
- Brand voice arbitrationWhere 'on brand' becomes 'not quite right' for reasons that aren't easily codified. A human still has to make that call.
- Client relationshipsThe conversation that earns the trust. The escalation when something goes wrong. The check-in that nobody pays you for.
- Compliance & legalWhen the cost of a wrong answer is regulatory exposure, a human signs off, and a documented chain of review backs them up.
- Hiring & cultureThe decisions about who joins, who leads, what gets celebrated. AI gets nowhere near these.
- Creative directionDirection stays human. Production, the variants, the executions, the volume, that's where AI earns its rent.
How an engagement runs
Audit → build → govern.
Audit & opportunity map
Two-week scan. Workflows reviewed against where AI is honestly useful versus where it's hype. We hand back a map of where to spend the budget and, just as importantly, where not to.
Build the working version
Pick the highest-leverage opportunity from the audit. Build the pipeline end-to-end, eval harness included. Ship it into the actual workflow. Measure what changed.
Govern, tune, expand
Cost monitoring, drift detection, prompt versioning, ongoing tuning. As the first build settles, the next one starts. Programmes scale; ad-hoc projects don't.
The stack
Whatever the job needs. We have no team-jersey allegiance.
When to talk to us
Briefs we actually want.
- → You're tired of AI proofs-of-concept that never make it into the business.
- → You have a workflow that's clearly mechanical and clearly burning hours.
- → You're spending more on a tool than the work it's supposed to be doing would cost.
- → Your catalogue has outgrown the content pipeline that supports it.
- → You want a copilot inside your own data, not another third-party SaaS subscription.
- → Your competitors haven't figured this out yet, and you'd like to keep it that way.
Where this plugs in
AI is the wiring inside almost everything else we run.
Don't know how you should use AI?
Neither do we, yet. But we know the shape of the work, where it ships and where it falls over. Tell us what your business actually does and we'll come back inside a day with a read on where AI would genuinely move a number.
Book an audit