2026 AI operating engagements · mid-market

AI that
works for you

Most AI gets bolted onto whatever your team is already doing. We work with you to put AI exactly where it genuinely adds value, with the rules and structure around it that make it actually work.

You are using Claude and ChatGPT already. Drafting documents, summarising meetings, getting first-pass answers through the day.

Useful. But what you are using is a fraction of what AI can actually do. The real lift comes when AI sits inside how the business runs, with proper context on the people, the work, and the goals.

We work with your senior team on where AI actually belongs, and how it supports where you are taking the business. The judgment that gets us there is honed across two decades of organisational design, strategy and process engineering inside operating businesses.

That is when AI becomes a superpower.

The position this firm operates from

Not just a brain.
A control plane.

What we install is a control plane that sits above the systems your team already uses. We call it Polaris. Its parts: a live Work Graph of how your business runs, the brain that decides, the interventions where AI genuinely adds value, and the measurement that proves it moved the number. The four pillars below are the brain. One part of the whole.

The Work Graph · the part you keep
A live, interactive model of how your business runs.
Every process, system, handoff, cost and bottleneck, mapped as one connected model you can open and explore. Not a slide, not a static PDF. The map we build to diagnose you is the same operating layer we leave running, and it stays live after we hand over.
The brain · four pillars working as one
01
Memory
Knows what to trust
Owns the business's hypotheses and commitments. Caches everything else with provenance. Never the system of record for facts that other systems own.
02
Sensitivity
Stays where it belongs
Sensitivity travels by lineage, not by tag. Anything derived from a sensitive source stays local. The cloud holds only operational data that has been cleared.
03
Autonomy
Learns when to act
Three tiers: propose only, act and notify, act silently inside a window. Trust is learned from your corrections, never self-scored. High-risk actions stay locked.
04
Reasoning
Convenes the council
For decisions that earn it, four AI systems from four labs debate in parallel. We call it Parallax. Disagreement stays visible. Most AI smooths over arguments. This one preserves them.

The brain is one part. Around it, the control plane scores where AI genuinely adds value, ranked by real ROI and acting only under permission, then measures every change against your baseline so we can prove it moved cost, speed, or revenue. Most teams build a piece of this in isolation. As far as we can see, no one else has put the whole combination together yet. The two genuinely differentiated parts: sensitivity by lineage rather than by tag, and the integrated setup itself. The working version runs in our own businesses every day. The version installed for a client is the same shape, fitted to your operating reality.

See it in action · a real run

Same question. Single AI versus Parallax.

Anthropic OpenAI xAI Qwen Synthesiser
The question
"Our pitch hit rate is excellent and we tripled revenue in two years, but pipeline keeps drying up. The team has the tools — CRM, outbound sequences, account frameworks — and almost nothing gets used consistently. I close most of the major deals myself. Our biggest vertical contracted sharply last quarter, leaving a $5M gap in 2026. What's the strategic move?"
— founder, mid-market experiential agency, ~60 staff
If you asked one AI
Build a unified client intelligence layer. Insight is the missing capability.

You've built strong delivery and a high-velocity team, but you're flying without a unifying intelligence layer. The single capability that could compound across pipeline, pitch, retention, and access is a proprietary client and category intelligence layer: a system that knows your accounts, their decision windows, their stakeholder maps, their competitive context, and surfaces the right next action at the right moment.

The structural argument: Most growth agencies hit this wall around 50 to 100 people. The founder's pattern recognition was the operating system; once the team is too big for that to scale, the answer is to externalise the founder's intuition into a system the team can run. AI-enabled client intelligence is the cleanest version of that today: ingestion of CRM, market signals, account news, and prior engagements, with structured outputs that prep account teams before every conversation.

Across the four jobs: Intelligence improves pipeline by surfacing leading indicators and prioritisation. It improves pitch by giving the team category-specific POV at parity with the founder. It improves retention by surfacing risk and next-best-conversation prompts. It opens access by enabling warm-feeling outreach and trigger-based introductions into new verticals where the firm lacks history.

The engagement shape: A four-week whole-business assessment to map the intelligence requirements, followed by a $100k to $250k build of the intelligence layer integrated into the team's existing workflow.

The founder problem solves itself. Once the team has the intelligence layer, they no longer need her instinct for context. She frees up. The dependency dissolves.

Confidence: 4/5. The intelligence layer is a well-validated pattern for agencies moving from founder-led to systemised revenue.

Scroll for the full answer
What Parallax produces
The intelligence layer is the wrong solution. This isn't an information problem dressed up as one.

The prior thesis is elegant and entirely wrong for this firm. Re-read what the founder actually said: the team already has the CRM, the outbound frameworks, the account ownership models. They don't use them. Leads go untracked. The founder closes more than the rest of the team combined. Adding a more sophisticated intelligence layer on top of a team that already fails to use basic tools is paying $250k to make the unused layer slightly fancier.

The real shape of the problem. Four constraints, mutually reinforcing but not the same thing:

  • Market concentration: the $5M gap from one vertical's contraction is a diversification problem.
  • Founder dependency: the founder is the sole functional path for the high-leverage revenue moments.
  • Sales-management failure: frameworks exist, no one enforces the cadence. There may be no functional revenue leader at all.
  • Access in new verticals: no relationship inventory, no proof, no procurement standing in the categories the firm needs to enter.

An intelligence layer addresses approximately zero of these as the lead intervention. It can help later, once the underlying operating model exists. It cannot replace the operating model.

What's actually needed, in order. First, a diagnostic that tests where the founder is the sole functional path, what the team would do without her, whether anyone holds the authority to enforce sales cadence, and which slice of the $5M is recoverable from retention and expansion versus new-logo acquisition. Second, a build phase, only if a named internal revenue owner with actual authority exists. Otherwise any system built will go unused, like the current ones. Third, AI workflows enter the picture as the leverage layer once accountability is in place. Not before.

The uncomfortable read. Most "intelligence layer" pitches to founder-led services firms are sophisticated-sounding non-solutions sold by advisors who can't, or won't, tell the founder the problem is human, not informational. The harder conversation, which is the Parallax position, is to name the management gap, name the founder behaviour change required, and sequence the work so the system gets installed only when the operating model will actually use it.

3/5 confidence · oscillation halt at round 3
Unresolved: retention vs new-wedge lead "weaponised POV" as access mechanism founder willingness to change behaviour
Scroll for the full answer
An actual Parallax run on a real client situation, anonymised · three rounds · four labs (Anthropic, OpenAI, xAI, Qwen) · honest oscillation halt rather than forced consensus
Read the architecture

Four phases. Fixed deliverables.
Roughly 11–14 weeks end to end.

01
Discovery2 weeks
Conversations with your senior team and the people who actually run the work. Mapping where AI could meaningfully change how the business operates, where it would genuinely add value, and where it would be a mistake. End-state: a written diagnostic you can act on whether or not we continue.
DeliverablesThe first cut of your Work Graph, plus a 20–30 page written diagnostic. Standalone-usable.
02
Architecture3–4 weeks
Designing where AI fits, and the rules around it. The voice leadership writes and judges in, captured so AI tools sound right. A clear map of what humans decide, what AI decides, what runs as a hybrid. The rules for when AI acts on its own. A rolling 90-day check on whether what we put in place is working. Documented so the system survives team turnover. Codified for retrieval, not just storage: most teams' first mistake is treating capture as the goal.
DeliverableThe full set of working documents.
03
Implementation4–5 weeks
Building the AI layer inside the tools your team already uses. Not new software to learn. Tool selection, integration, security and governance done in the background. The design becomes a system that runs.
DeliverableWorking control plane.
04
Calibration2–3 weeks + optional retainer
Live calibration against the first weeks of real use. Where the system fits, where it does not, what needs fixing. Then handover: your team runs it from here, with an optional retainer if you want a second set of eyes through the next quarter.
DeliverableCalibrated system + handover.

Better outcomes.
Faster. Cheaper.

A live Work Graph of your business you can open and explore, plus the working documents below that take the recurring thinking work off your team and lift the standard of every decision. Not deck slides. Not theory.

01
The way leadership writes and decides, codified so every AI tool in the company sounds and judges like you do.
02
A clear map of who decides what. What stays with people, what runs on AI, what runs as a hybrid.
03
A record of how past decisions actually landed. So the AI integration learns from what your business has actually done, not just from average data.
04
A rolling 90-day check on the quarter. A real read on whether it worked, not a wishful OKR cycle.
05
AI built into the tools your team already uses. Not new software to learn.
06
Clear rules for when AI acts on its own. When it asks first, and when it gets out of the way.
07
Context the AI always sees. So it is useful in every conversation, not just the ones where someone primes it first.

By day 14 you have a live map of your business in your hands.
Not a slide deck. No lock-in.

D 01–03
We talk.
Two structured conversations of roughly two hours each. Senior team, executive sponsor, operating context, current pain. I leave with a clear read on whether the engagement will work, or whether to walk.
D 04–10
Discovery interviews.
Six to ten people in your senior layer plus the operators who actually know how work gets done. Each interview is roughly an hour. I read transcripts overnight. Patterns surface fast.
D 11–14
First-cut diagnostic.
The first cut of your Work Graph, plus a 20–30 page written diagnostic. We meet to walk through it. You push back where you disagree. I revise. If you stop here, you still have a Phase 1 deliverable that is useful on its own.

Indicative ranges, measured against
a pre-engagement baseline.

Every intervention is tracked against that baseline, so we can show whether it actually moved cost, speed, or revenue. That is also how we prefer to price: on what moved, not on hours billed.

Featured outcome
3050%
Reduction in internal reporting and briefing cycle time
The single largest unit of cognitive overhead in most mid-market companies. AI does not eliminate it. It compresses it to the residual judgment layer that actually requires a human. In my own businesses the headcount layer has compressed materially further than this. See [02] Track record.
Range consistent with published findings from McKinsey (The State of AI in 2025: Agents, Innovation, and Transformation, Nov 2025), Microsoft (2025 Work Trend Index: AI at Work Is Here. Now Comes the Hard Part, Apr 2025), Anthropic (Economic Index Report, Jan & Mar 2026), and Stanford HAI (AI Index Report 2025).
Decision velocity
Faster decision turnaround on recurring decisions
Investment committee, hiring, vendor selection, customer escalation. The recurring lanes get codified.
Coordination
Less alignment overhead at the exec layer
Fewer meetings to align. More to decide.
Key-person risk
Institutional memory captured
Quality
Decision-quality measurable
Audited against your own pre-engagement decision baseline.
Recurring cognitive work
Document drafting, meeting summarisation, follow-up tracking, research synthesis
The work knowledge teams spend most of their week doing. Re-architected so AI carries the load and humans carry the judgment.

Three tiers. Fixed-fee per phase.
Quoted against scope.

Tier 01
Diagnostic
2 weeks
Phase 1 standalone. For companies that want the assessment before committing to the architecture.
  • Discovery interviews
  • Workflow + decision-pattern audit
  • Written diagnostic, 20–30 pages
  • Phased proposal for what comes next
Tier 03
Full operating engagement
11–14 weeks
All four phases. Exit with a working control plane, calibrated against real use, with documentation and ongoing-retainer optionality.
  • Everything in Architecture
  • AI built into the tools your team already uses
  • Integration, security and governance
  • Calibration against the first weeks of real use
  • Optional retainer through the next quarter

Writing on AI.

14 Jun 2026 · Featured
The personal AI operating system, four pillars, one brain.
The architecture we install for clients, written up after we'd settled it by multi-lab debate and pressure-tested it against the field. Memory that knows what to trust. Sensitivity that travels by lineage. Bounded autonomy that learns from your corrections. Reasoning convened only when a decision earns it. Two pillars with no verified prior art, and the diagram that ties them together.
22 Jun 2026
Your AI is reading your confidential data. The real question is where it goes.
Sensitive data is already flowing into AI in every department, and most companies cannot say where it ends up. The four very different answers to "where does my data go," the two questions that decide everything, and why the first move costs nothing in hardware.
26 May 2026
I replaced an eight-person team with AI.
At Kasu, a four-person engineering team became one AI-augmented CTO. A four-person marketing team became an AI workflow. Same business, same output, a fraction of the cost. The methodology, the math, and the 2025 research showing only six percent of companies are getting AI right.
04 May 2026
How to start an AI argument.
Subjecting AI outputs to iterative critique from competing models produces better results than accepting the first answer. Productive disagreement embedded into the AI development process itself.
22 Apr 2026
Get your own AI team.
Inside a multi-agent AI system. A primary chief-of-staff coordinator manages eight specialist agents, each with their own role, voice, and area of authority.
26 Feb 2026
The AI crisis.
South Australia is building elite advanced-manufacturing capability while ignoring the mass displacement of mid-tier knowledge workers already underway through AI automation.

Hired short.
Kept on long.

The methodology was built inside operating businesses, not from outside them. The pattern that produced it shows up across two decades of real work.

Operator history2002–2017
Hired as a consultant. Kept on as an operator. ClarkMorgan, DMG Entertainment, the Australian Pavilion at the Shanghai World Expo, IVG, OSP Holdings, TenTen Entertainment. Each engagement started short. Each ended with an operating role offered because the outcomes were good enough that the client decided the work belonged inside. The methodology gets honed inside the business, not from outside it.
Faculty GroupFounding partner
One of four founding partners. Scaled the marketing agency, the incubator, the accelerator and the crypto launches from start-up to 150 people across eight businesses. Designed and shipped a custom multi-LLM interface for customer-archetype generation eighteen months before this was a category. The build proved the thesis.
Kasu2025
Replaced a four-person development team with one AI-augmented CTO. Replaced a four-person marketing team with an AI workflow. Same output. Fraction of the cost. Currently running in production.
Portfolio2026
Twenty-plus active projects executed as a sole operator. AI is the team.

Built and operates the working
operating system this methodology comes from.

Luke
Lombe.
Principal · Echelon One

Twenty years of being hired short and kept on. ClarkMorgan, DMG Entertainment, the Australian Pavilion at the Shanghai World Expo, IVG, OSP Holdings, TenTen Entertainment. Each engagement started as a short consultancy. Each ended with an operating role offered because the outcomes were good enough that the client decided the work belonged inside. Then Faculty Group as one of four founding partners, scaling marketing, incubation, acceleration and crypto from start-up to 150 people across eight businesses.

The disciplines: business process engineering, strategy, facilitation, organisational design, behavioural psychology and organisational dynamics. Honed inside operating businesses, not from a deck.

The methodology Echelon One installs is the same one behind the eight-person team math at Kasu, the multi-LLM AI work at Faculty Group, and the twenty-plus-project sole-operator portfolio Luke runs today. Refined through actual operating pressure. The four-pillar architecture is documented, and the working version is in use across our own businesses every day.

The differentiator is not the slides. It is that the working version exists. You can see it. Then you walk through what the corporate version looks like in your business.