By Ben Long — Co-Founder & CTO

There’s a moment in every engineer’s life where you realise the machines aren’t misbehaving — they’re simply reflecting the chaos of the world you asked them to interpret. When we started building Flambé Optimiser, that chaos arrived in buckets. Menu structures that looked like they’d been written during a power outage. Suburbs with naming conventions that defy cartography. Plugins that output JSON with all the grace of a dropped lasagne. And then the AIs themselves — each model with its own quirks, biases, blind spots, and preferred ways of interpreting reality.

People imagine AI as a single, monolithic intelligence. In practice, it’s closer to a room full of bright but unruly interns, each brilliant in one area and hopeless in another. When you ask them all the same question, you often get three different answers and a philosophical debate.

My job — our job — was to turn that into something useful.

That’s where context coding emerged: not as a theoretical framework, but as the practical necessity of building an AI-optimisation engine that could survive in the wild. If traditional SEO was about interpreting search engines, context coding is about wrangling the machines themselves and shaping their understanding of the world.

And the F&B world, as I quickly learned, is one of the messiest.

Revolting AIs and the reality of competitive interpretation

When we first began testing AI models on F&B websites, the results were… enlightening. One model was brilliant at reading menus but terrible with suburb names. Another understood cuisine classification but treated dietary labels like decorative garnish. A third could parse JSON-LD schema but misinterpreted opening hours if the formatting deviated from a narrow norm.

Most people would pick one model and try to mould everything around it.
We took the opposite approach.

We embraced the revolt.

We created a competitive structure, a multi-model orchestration layer where different AIs evaluate each other’s interpretations. If one model misreads something, another challenges it. When two models disagree, we run adjudication passes. When three models disagree — which happens more often than you’d think — we analyse the edges until we understand why.

In other words: we built an environment where AIs correct each other before they ever output a score for a customer.

It’s noisy, complex, and occasionally comedic. But it works.

Why GEO is the battlefield — and why static rules won’t cut it

Local search is evolving faster than any SEO trend we’ve seen in 20 years.
Models continuously update their understanding of:

  • what counts as “nearby”
  • which cuisines trend in which suburbs
  • how people phrase local queries
  • which attributes matter most in a recommendation
  • how intent shifts across hours of the day or season

A static rule set is dead on arrival.
A quarterly update cycle is slow motion.
A human-curated rulebook is impossible.

So we built Flambé’s context engine to do something different:
continuously ingest, compare, and adjust to the way AI models themselves evolve.

Where most tools scrape keywords, we scrape behaviour — shifts in how models interpret real data. We built pipelines that track changes in:

  • local intent patterns
  • menu interpretation logic
  • entity-relationship weightings
  • context relevance scoring
  • AI preference drift

When a model changes how it values “open now” versus “top rated,” we see it.
When suburbs begin clustering differently in AI understanding, we map it.
When dietary labels suddenly matter more, we adjust.

This is how we stay ahead of the curve on GEO — by treating AI not as a fixed input but as a moving landscape. A search horizon in motion.

The single updating, trend-aware rule aggregator

At the heart of Flambé is something we haven’t talked much about publicly:
our rule-aggregation engine.

It sits quietly behind the product, continuously merging:

  • what the AIs believe
  • what the web signals show
  • what the local trends indicate
  • what the data quality suggests
  • what the models are learning from each other

Most SEO tools run an audit.
Flambé runs an interpretation negotiation.

The system doesn’t just check your website — it checks how the current generation of AI models understand your website. And when those models shift, so does the interpretation.

This is why Flambé’s scoring feels more “alive” than traditional audits.
Because it is.

The advantage for our users

The result of all this orchestration, adjudication, and rule-aggregation work is simple for our users:
You stay visible even as AI search rules evolve.

You don’t wake up one morning to discover your bookings evaporated because a model changed how it evaluates local businesses. You don’t rely on outdated SEO principles that assume ranking is static. You don’t disappear from voice-driven “What’s good near me?” queries because your menu wasn’t machine-readable in the new format.

Flambé is built to evolve at the speed of AI — because we architected it with the expectation that nothing would stay still.

Why I believe context coding is the next great business advantage

After decades in engineering leadership, I’ve learned that technology always rewards those who build for the world as it will be, not the world as it was. Context coding is our way of giving every F&B business — from family cafés to multi-venue operators — access to the same AI-awareness advantage the big players use.

It’s not glamour work. It’s not a marketing tagline.
It’s infrastructure.
It’s stewardship.
It’s competitive resilience.

And in an era where AI controls the front door of customer discovery, resilience might be the most valuable asset of all.