Services

Five engagements.
One principle.

We leave when your team can run what we built without us.

Most AI consultancies sell dependency. Retainers that compound. Systems only they can maintain. We do the opposite. Every engagement is scoped around a handoff from day one. Here's how.

Which One

Start small. Ship. Then go wider.

Most teams come to us with one of these questions.

"We don't know what to do first."
The Diagnostic
"We know what to build. Build it properly."
The Sprint
"We need to ship consistently across the business."
The Accelerator
"We want our own people building this."
The Program
"Our Dust bill is growing faster than the work."
The Cost Optimisation Audit
Engagement 01

The Diagnostic

2 weeks · Fixed price · For leadership teams

You're being pitched AI from every direction — vendors, internal champions, the board, LinkedIn. You need someone to sort signal from noise against your actual business.

That's what this does.

Phase 1 · Week 1

We map the real work.

Interviews and workflow shadowing across the teams you care about. How work actually moves, not how the org chart says it does. Where handoffs stall. Where hidden manual effort lives. What a 10% gain means in your P&L vs. where it doesn't matter.

Phase 2 · Week 2

We build the business case.

Every opportunity scored on reach, impact, effort, dependencies, and risk. Build vs. buy vs. hybrid call per project. Sequenced so later projects get cheaper, not more expensive.

You leave with
  • Quantified operational baseline tied to real data
  • Prioritized roadmap: build this, buy this, skip this
  • Business case behind every call
  • 90-minute findings session with your leadership team

Previously scoped: 10-day on-site audit across 6 departments at a global OTA. 40 workflows mapped. 6–8 sequenced AI projects. Roadmap targeting 30% profitability uplift. Read the case study

Engagement 02

The Sprint

4 weeks · Fixed price · For teams with a clear problem

You already know what to build. You don't need another audit. You need someone who's shipped these systems before and will build this one properly the first time.

What "properly" means.

Not a prototype. A system designed for production from day one: evals so you can tell if the output is good, guardrails for when it isn't, observability for when it breaks, documentation for the person who inherits it.

You leave with
  • One production-ready workflow, deployed
  • Evals and monitoring in place
  • Team training and documentation
  • 30 days of post-launch support

Previously shipped: a voice agent that freed 30% of sales time at CitySupply. A fully-automated outbound engine driving 4× faster sales cycles at BaseClaims. An onboarding agent that cut onboarding time by 95% at Regulars.

Engagement 03

The Accelerator

Monthly retainer · For organizations ready to build at pace

One shipped workflow is not a transformation. You need to ship consistently, across multiple teams, without dependency on outside help.

Three tiers, depending on scope.

Focused

One revenue lane — Growth, Sales, or Success.

  • 1 production workflow shipped per month
  • Bi-weekly strategy calls
  • Shared Slack channel
  • Monthly performance report
  • SOPs and async team training

Best for: teams proving AI in one area before going wider.

Book an intro call
Embedded

Unlimited delivery, one embedded engineer.

  • Unlimited workflow delivery
  • Full-time forward-deployed engineer
  • Weekly executive sync
  • Priority support (4hr response)
  • Custom AI tooling for your stack
  • Quarterly roadmap planning

Best for: organizations treating AI as a strategic advantage, not a line item.

Book an intro call
Engagement 04

The Program

6 weeks · 10 modules · Up to 6 participants · On-site or remote

Every engagement above ends with a handoff to your team. This one is the handoff, sold on its own.

Six weeks that take a cohort from opening a terminal to shipping agentic systems in production, with the judgment to know what's worth building in the first place. It works for two kinds of people: business and product folks who want to build for Growth, Sales and Ops without waiting on an engineering queue, and engineers who want current patterns for agentic systems and how to embed them in the company.

01 · Every week

Something ships.

Every module ends with something running: an agent, a workflow, a deployed system. Exercises are built on your industry and your processes, so week-one output is already relevant. By week six the cohort has closed with a real internal workflow in production.

02 · Across six weeks

The muscle compounds.

Repetition builds instinct. After six weeks of building, your people can take a new idea, break it into steps, and build it themselves, long after the program ends.

03 · Throughout

ROI discipline.

A repeatable method to find the highest-return workflows, prioritise them, and kill the ones that won't pay. AI projects, not AI theatre.

Ten modules
  • LLM & RAG fundamentals — model choices, embeddings, chunking, retrieval tradeoffs
  • Context engineering — system prompts, tool descriptions, structured outputs, memory design
  • Knowledge bases & memory — structured vs. vector, memory that persists
  • Event-driven workflows — webhooks, CRON, tool calling, hooks
  • Full-stack agents — async jobs, streaming, websockets, agentic UI patterns
  • Evals & quality control — judge loops, judge calibration, versioned eval sets, regression testing
  • Human-in-the-loop — certainty-based routing: accept, reject, escalate
  • Deployment & monitoring — traceability, telemetry, prompt versioning, drift detection
  • AI-first development — MCPs and the Claude Code harness: memory, hooks, custom commands
  • Scoping & AI roadmap — 5-step diagnosis, RICE prioritisation, sequencing, governance

Before the program, AI felt like an alien thing. Now, when I have an idea, I feel like I can sit down, break it into the right steps, and build it.

Olympia · Program graduate

Engagement 05

The Cost Optimisation Audit

Fixed price · Remote · For Dust workspaces

Dust is moving from tier-based billing to usage-based. Under tier pricing, a wasteful agent and an efficient one cost you the same. Usage-based billing prices the difference.

Most of the waste sits in the plumbing. A sync agent re-copying rows a script already computed. An afternoon briefing that recomputes what the morning one already said. A sub-agent returning 14,000 tokens that its parent re-reads at every later step. Nobody built any of it badly. It grew, and nothing was watching.

100+Agents diagnosed

One workspace, 35 users, every agent opened and costed.

−65%Spend reduction

Identified, sequenced, and costed, with an evidence grade on every number.

Output per credit

More agent work for the same budget, against the pre-audit pace.

01 · Diagnose

The whole agent pool.

Every agent, its real cost, its real usage, its owner, its value rating, reconciled against your billing so the baseline matches what you were actually charged. Includes a trigger-ownership census, which is the only workspace-wide view of your schedules, since Dust triggers are visible solely to whoever created them.

02 · Dissect

The ten workflows that carry your spend.

Opened up from their real runs. One page each: what it does, where the money actually goes step by step, and what to change, ranked by effort class — a settings toggle, an instruction paste, or a small rebuild.

03 · Generalise

The optimisation playbook.

Eight plays that generalise past your top ten, each with the trigger that tells your builders when it applies, plus a reusable template for workflow #11 onwards. This is the durable part.

04 · Watch

Thirty minutes a month, automated.

Scripts you run monthly: spend reconciliation, spike attribution, trigger census, per-tool cost and error drift, zero-usage sweep, and two automated watches that surface new optimisation candidates before they become line items.

05 · Govern

A checklist before anything ships.

Bounded queues with an "already tried" marker. Named owners. Return contracts on sub-agents. The paid-twice check: does a platform you already own do this natively. No automation running on personal credentials. And a validation protocol for every model or instruction change.

You leave with
  • A usage baseline reconciled against your billing
  • Ten workflow dossiers, ranked, costed, and graded by evidence strength
  • The optimisation playbook and the template for the next workflow
  • Paste-ready instruction redlines and a model-fit review
  • Monthly review scripts, handed over and run once together
  • Two governance checklists: build-time and monthly admin
  • A rollout tracker with owners, expected saving, and a realised-savings column

Vlad dove deep into how we were using Dust⁠—identifying what we were doing wrong, what could be improved, and opportunities we hadn’t considered.

Felix Keser · CEO, Nexio Projects

What that engagement delivered: up to 95% cost reduction on the priority AI workflows, credit spend on a major sync taken to zero from 1,210 per run, and a 40% cut in token usage across agent instructions.

The Thing Nobody Else Says

Every engagement ends with a handoff.

Not "ongoing partnership." Not "trusted long-term advisor." A handoff.

On the Sprint, it's the team that'll run the workflow we just shipped.
On the Accelerator, it's the team we've been training alongside us for months.
On the Diagnostic, it's the internal leader who'll own the roadmap.
On the Program, the handoff is the whole product.
On the Cost Audit, it's the admin who runs the monthly review from then on.

The test we apply: can the team run this without us? If the answer is yes, we step back. If the answer is no, we haven't finished.

This is the single biggest reason to work with us instead of hiring a freelancer or signing a long-term consulting retainer. Both of those are optimized for you to keep paying. We're optimized for you to stop.

Not sure which one you need?

That's what the 30-minute call is for.