← AstraJax

Airspace LA · Matthew's journey

From actor to architect — without writing code.

The set-up

Airspace keynote · 2026

From actor to architect — without writing code.

The talk track from Matthew Hopkinson's Airspace keynote: how a non-technical commercial leader turned messy field sales operations into an AI-ready operating system — boring layer first, agents second.

  1. 01The boring layer — cleaning up the operating system.
  2. 02The agent layer — what became possible once the data was clean.
  3. 03The lesson — what this means for teams trying to use AI properly.

I

No code, no clue — but plenty of context

Matthew spent the first ten years of his career as an actor — trained…

Act I · 1

Employee number 33

Butternut Box hired Matthew to lead their first face-to-face sales team out of London — an offline growth channel with a cost per acquisition that wasn't at the mercy of digital.

Twenty-six-year-old Matthew looked them dead in the eye, completely clueless, and replied: "I can deliver that." Faking it for the next two years was pretty fun. Anywhere they thought there might be a dog, they'd go. The Birmingham Knitting Fair was a personal favourite — and it brought in the highest single-day sales-per-shift in history.

Neanderthal Matthew in Hulu & MGM's 'Dawn'

Seed fund to unicorn
Act I · 2

Scale catches up

Cut to 2026: the scrappy startup has gone from 33 employees to 1,500 across seven international markets. Small seed fund to unicorn status.

II

Brilliant people, low-leverage tools

They grew fast

Tools that never talked
Act II · 1

The operational weight

1,500 activations a year with 120 salespeople attending them weekly. An office team of 15.

Direct Sales is full of brilliant people, doing high-value work with low-leverage tools.
The 10M-cell breaking point
Act II · 2

The Google Sheets problem statement

Cracks were showing — team progression slowing, burnout a real risk, job satisfaction tanking. It wasn't just scale. It was the visibility of what was possible with AI, and the realisation there was no clear route to realise those advantages.

III

The boring layer

Over the next twelve months, they built the foundation: clean, centra…

From inboxes to an operating system
Act III · 1

Gmail, WhatsApp, Notion and Sheets → an operating system

The agents, custom interfaces, Slack bots and AI workflows all came later — and only moved quickly because the boring layer was already there.

We went from Gmail, WhatsApp, Notion and Google Sheets… to an operating system.
The event interface, after
Act III · 2

Events — from spreadsheet rows to role-specific interfaces

Before: each row was an event — dog show, garden centre, shopping centre — with dates, costs, staffing, logistics, forecasts, organiser comms and spend attached. The sheet technically contained information, but it didn't create intelligence. It didn't know who needed what.

Staffing dashboard + lock
Act III · 3

Staffing — four hours to a structured recommendation

Before: a Regional Manager manually assigned people across availability, training, driving, location, shift load and development — then screenshot the rota into WhatsApp. One week of staffing could take about four hours.

Forecasting + the labour-model unlock
Act III · 4

Forecasting — when the average was lying

Before: forecasting relied on people knowing which events felt good, which regions were strong, which assumptions were probably wrong. It doesn't work at hundreds of activations across multiple regions, labour models and spend profiles.

The old system could tell us what happened. The new system can guide us in making the best decisions.
What the foundation unlocked
Act III · 5

What the foundation unlocked

Roughly £180k a year saved through better travel planning.

IV

The agent layer — two weeks, not two years

Twelve months building the boring layer

Targeted agents, narrow scopes
Act IV · 1

Targeted agents, not one magical assistant

If you try to add agents on top of messy data, they become very confident chaos machines — fun, but not very useful.

That speed was only possible because the foundation was already there.
Clive — personality as adoption
Act IV · 2

Clive Wigglesworth Esq

Victorian gentleman. Emotionally needy. Desperate for approval. Furious that he needs it. Modelled partly after Matthew's golden retriever Ajax.

I think I fancy him.
Reggie runs the bonuses
Act IV · 3

Reggie — the oblivious uncle

Reg looks after bonuses and payroll QA. The fleet's beloved oblivious uncle, doling out coins and warm wheezy laughter while entirely missing the soap opera around him.

Tashi · Marlowe · human · Marcel
Act IV · 4

The Trinity

The cleanest example of agent orchestration: organiser emails. Every event organiser communicates differently — PDFs, invoices, logistics packs, or the important bit buried in paragraph seven.

Tashi links. Marlowe proposes. The human approves. Marcel executes.
It grew arms and legs
Act IV · 5

Arms and legs

There's a running joke with Matthew's MD. Every week she'd ask how the systems were going. Every week he'd say: "It's grown arms and legs." Eventually she'd just say "Arms and legs?" and he'd nod.

V

Trust, training, value, safety

AI adoption is not just a technical journey

Act V · 1

Trust

Clear outputs, clear explanations, visible audit trails. Agents with narrow jobs — and moments of lightness.

The personalities mattered. The jokes mattered. The confetti mattered. The fact the team could mock the bots mattered. It turned AI from something happening to them into something they could play with.

Act V · 2

Training

If people don't know how to use a system, they don't experience value. If they don't experience value, they don't feel safe. And if they don't feel safe, they disengage.

When people don't get good outputs from AI, they often assume the model is bad. Often the issue is they haven't yet learned how to communicate the work clearly. That is now a management skill — not a technical skill.

Act V · 3

Value

People need to feel the system makes their work better — not just faster.

If a Regional Manager spends less time building a report and more time coaching their team, that's value. If an Event Coordinator spends less time hunting organiser emails and more time improving event quality, that's value. You have to shout about it.

The close
Act V · 4

Safety

A lot of the manual tasks people are comfortable with will disappear. That can be exciting. It's more often frightening.

The agents can take the sludge. The humans keep the meaning.