Adoption
78%
▲ 71 pts62 of 79 eng
The published methodology
The Augmented Company Blueprint
The complete, role-by-role picture of a company that works augmented. Your agentic setup is your company’s new asset—built by your own people bottom-up, measured like production software, and governed like one. This is the standard we train and certify against.
The argument, in under a minute
One person is linear. A company is the multiplier.
The whole blueprint before the detail. Everything further down this page is the evidence for these seven steps.
Figures marked ✱ on this page are illustrative — they show how the model behaves, not results we are claiming. Every one of them is replaced by measured client data or shown as nothing at all.
The argument, written out
- 1
One person builds it
A real tool, built on their own work. Real hours come back — and they come back to one desk.
- 2
The bottleneck just moves
A local gain becomes a business gain only when the next constraint can absorb it. Usage rises; delivery does not.
- 3
Sharing is the multiplier
Not a stronger model — the same capability reaching more people, without losing ownership or control.
- 4
Which is why it has to be bottom-up
Covering the first 80% of a role’s work is fast. The rest is slow — and it is exactly where a generic top-down tool starts to drift.
- 5
Nothing spreads without proof
Personal, colleague, team, company — every step has a gate. Risk levels and an approval floor keep a hundred agents from becoming chaos.
- 6
Measured, not claimed
No baseline, no measurement — only assertion. Capability telemetry joined to your own delivery systems makes a claim a buying committee can inspect.
- 7
That is an Augmented Company
Five layers, live at once. The harness underneath is swappable — the asset your people built is not.
Evidence before architecture
What this looked like at Enersis
Enersis ran the program across engineering, HR, and operations.
Across functions
The program included people in engineering, HR, and operations.
Development cycle
CTO & Managing Director Florian Wolf reported that one developer shortened his development cycle ‘by a factor of seven or so.’
From internal tool to customer work
Florian Wolf also described an internal chatbot that progressed into Enersis’s product and customer MVP work.
These are client-reported testimonial observations. They are not audited ROI and do not establish a general performance expectation.
See the full Enersis storyMeasurement that survives review
For new engagements, we establish the baseline before work begins and test later claims against the client’s delivery systems. Usage alone is never labeled productivity.
Delivered by practitioners
Adam Bilišič
12+ years of professional experience; former CTO; more than 8 years leading teams for enterprise clients.
Jerguš Frajt
10+ years as a full-stack engineer delivering complex systems for Swiss enterprise clients.
The whole system at a glance
Five layers. Each one is real work — and each one is covered.
Most AI initiatives build one layer and wonder why nothing compounds. Open a layer to see how it works.
Layer 1the foundationThe foundation is swappable — the asset is not
Works on your stack. Survives your next stack.
Everything is built on open standards — Git, OpenTelemetry, open skill formats, MCP. Pick your harness below and watch what changes: one layer.
GovernanceStays
AOF — Agentic Operations Framework
Ownership, versioning, risk levels, budgets, promotion gates
ObservabilityStays
OpenTelemetry + Langfuse
Performance · cost · quality, per team — vendor-neutral standards
Your assetStays
Orchestrators · skills · micro-tools · MCP servers · memory
What your people build and what the system learns — held in files, versioned in Git, portable by design
Source of truthStays
Git
Everything versioned, reviewed, auditable — the industry standard
HarnessSwappable
Running on Claude Code / Claude Enterprise
Currently the strongest harness for this work — our reference setup.
What a swap actually costs: Configuration migrates by script; behaviour gets re-validated — a different model responds slightly differently to identical setup. The method and the artifacts survive. Mileage varies by harness and model quality; nothing you build is stranded. Reference setup: everything above runs as-is.
Layer 2the engineThe simulator — every role, concretely
Check what each role automates. Watch it compound.
Open a role, select the work its people would automate, and set how many people hold that role. The panel shows the monthly capacity returned; a section further down shows how it compounds from person to team to company. Every figure is illustrative.
Start from a preset
Industry
Company size
Pick the closest match — it fills in a typical headcount for every role below. Change any of them afterwards; the numbers follow.
Share of the company augmented
0%
0 of 13 peopleAugmented Company at 60%
ppl0 h
ppl0 h
ppl0 h
ppl0 h
ppl0 h
ppl0 h
ppl0 h
ppl0 h
ppl0 h
ppl0 h
ppl0 h
Why pilots stall
A local gain becomes a business gain only when the next constraint can absorb it.
Usage can rise while delivery remains unchanged if capability stops with its original builder or moves the bottleneck elsewhere.
Delivery chain
- 1
Engineering
Execution accelerates
- 2
Product
Specification becomes the constraint
- 3
Commercial and onboarding
Adoption capacity becomes the constraint
- 4
Customer
Value arrives only if the whole chain moves
Sharing path
- 01One builder proves it
- 02A role reuses it
- 03Adjacent roles adapt the pattern
- 04The company governs it
The multiplier is not a more powerful model. It is a useful capability moving from person to colleague to team to company without losing ownership or control.
Where the investment actually returns
One builder is linear. Sharing is the multiplier.
The same capability compounds as it moves from its builder, through their team, and across the company.
Estimate · Illustrative default
×112 hper month
One person builds it
An orchestrator or micro-tool, built on their own work — encapsulating their know-how. Real, but linear: it saves one person's hours.
× team142 hper month
The role shares it
Through the marketplace, the same tool serves everyone who holds that role. The build cost was paid once; the return multiplies by headcount.
× roles318 hper month
Patterns cross roles
The PM's status synthesizer becomes the sales team's pipeline digest. Patterns, skills and guardrails transfer — each role starts further ahead than the last.
One person vs company — 12-month projection
Shared across company · 4,509 h/yrOne person · 132 h/yr
And this is why it must be bottom-up. Automation only works when it encapsulates the know-how of the person who actually does the work — every company's invoices, tickets, deploys and deals are different, and a generic top-down tool covers the commodity part while missing the part that is your firm. There's a second reason: getting from 0 to 80% coverage of a role's work is fast; 80 to 95% is slow and hard — and the uncovered slice is exactly where a fully autonomous "finished system" starts to drift, producing output that serves nothing. People who orchestrate, audit and extend their own systems are what closes that gap. So capability grows person → colleague → team → company — the multiplier lives in the sharing step, and the next section is what keeps it safe once it spreads.
Layer 3the memoryGovernance — AOF
What stops a hundred agents becoming chaos
The Agentic Operations Framework (built in the open, adopted as part of certification) governs how capability spreads and how it's kept safe.
The promotion path — nothing spreads without proof
Personal
built on your own work
gate: it ran, verified, boundedA colleague
second user, real feedback
gate: measured help, docs existThe team
shared, versioned, owned
gate: perf · cost · quality passThe company
marketplace, autonomous agents on Slack / CI / ticketing
gate: risk level + approval floor setThe same ladder carries what the system knows
A skill and a memory file are the same kind of asset. Both start on one person's machine, both are worth nothing until a second person can use them, and both travel these gates.
Personal memory
How you work. Your corrections, your preferences, the context you would otherwise retype every session. Lives on your machine and travels with you between projects.
one owner · nothing to reviewTeam memory
How this codebase works. Conventions, architecture decisions, the deploy path, what broke last time. Committed to the repo, reviewed in the pull request, inherited by everyone who clones it.
in Git · reviewed like codeCompany memory
What holds across every team. Security boundaries, approved tooling, the standards a new project starts with. Published once, versioned, inherited everywhere.
owned · versioned · inside the perimeterVendor-managed memory does not survive a harness swap. File-based memory in Git does — which is why it sits in Your asset, one layer above the harness.
Risk levels & the approval floor
Every agent action carries a risk level. Read-only digests run free; anything touching money, personal data or production sits above the approval floor — a human signs off, always.
Evidence is a by-product of running, not paperwork.
Budgets as blast radius
Token budgets per team aren't just cost control — they bound what any system can do before a human looks. Ownership and versioning mean every orchestrator has a name on it and a history behind it. Maintenance is scheduled: what stops earning its place gets retired.
Layer 4the perimeter
Security answered before it's asked
What reaches a model
Defined per risk level: anonymization on by default for personal and financial data; classification rules your compliance team signs off — set once, enforced by hooks, not by memory.
Where agents run
On your machines and inside your network — laptops, IDEs, your own CI runners and cloud accounts. Agents act through named, scoped service identities rather than a person's credentials, so every run has an owner and a blast radius set in advance.
Where models run
Enterprise cloud (Claude Enterprise, Azure / OpenAI, Google Cloud) or fully self-hosted models for the strictest data classes. The blueprint works on all of them — see the harness layer.
GDPR & audit
Telemetry doubles as the audit trail: what ran, on what data class, approved by whom.
Specifics are set per client policy.
Four quarters, one shield — together they are your AI security, complete.
Layer 5the resultMeasured, not claimed
The board sees numbers, quarterly
Telemetry proves adoption, capability quality and cost — precisely. It does not, on its own, prove productivity. So the figures above are our estimate, and this is how we would check whether it actually happened.
Delivery Impact Console
Spend / eng / mo
$186
▲ $24Q1 $162
Cycle time p85
4.1d
▼ 21%Q1 5.2
Change failure rate
⚠ WATCH11%
▲ 1 ptQ1 10
Rollout & response
ADOPTION vs CYCLE TIME · BY COHORTWave 1 — 4 teamsWave 2 — 3 teams
Adoption %
Cycle time p85 (days)
Wave 2 holds flat through Wave 1's improvement and bends only after its own June rollout — the staggered schedule acts as a control group.
The path — and the destination
From first workshop to Augmented Company
Companies progress through the stages of the published maturity model. Individuals progress through the level ladder inside it. One standard, both scales.
Stage 1
Workshop
Individuals prompt; nothing is captured
Stage 2
Cohort
Setups exist and are used on real work
Stage 3
Sharing
Capability shared across the team, measured
Stage 4
Governance
Governed at company scale — AOF holds
Destination
Certification availableAugmented Company
All layers live, measured, governed — the certifiable state
Certification, in practice: the methodology is published and both individual and company certification are open against it — the exams test what people built and apply, not what they can answer. Verification of every issued certificate is public. We built it rigorous before we launched it.
Where does your company stand?
Check what you already have
0 / 8 — most companies start here.
Free PDF Guide
Take the methodology with you
Get the published guide to the five stages of AI-augmented coding mastery — the ladder this blueprint is built on. The checklist above stays free either way.
Executive FAQ
Is this an engineering methodology or a company operating model?
It is a company operating model whose current entry point is usually engineering. It covers how capabilities are built, shared, measured, controlled, and eventually adapted across roles.
What can Augmented Club deliver today, and what is still on the roadmap?
Today we offer facilitated engineering-team workshops, an applied cohort, and both individual and company certification against the published methodology. Wider cross-role expansion is the blueprint destination and is scoped only after initial capability proves useful, owned, and measured.
What time and access do you need from our team?
A useful engagement needs an executive sponsor, participants applying the method to real work, baseline delivery data, security and compliance input, and named owners for anything promoted beyond personal use.
How do you distinguish AI usage from productivity and financial return?
Usage telemetry is leading evidence only. Productivity or financial claims require a baseline and evidence from the client’s delivery systems. Measured results, modeled capacity, and testimonial observations remain explicitly separate.
What data can reach a model, and can this run within our architecture?
The client defines data classes, permitted runtimes, anonymization, and approval rules. The deployment boundary is selected during scoping and may include approved enterprise-cloud or self-hosted infrastructure. Provider behavior, residency, retention, and contract terms must still be validated.
What survives if we change model or tool vendors?
Versioned skills, workflows, controls, and open integrations are intended to remain company assets. Tool-specific configuration may need migration, and behavior, security controls, and quality must be revalidated. Portability does not mean a zero-cost switch.
What is the smallest useful place to start?
One sponsored engineering team, one real workflow, baseline data, a named capability owner, and a clear control boundary. Expand only after that first capability is useful and measurable.
Three ways in