Pilot in command
One person is accountable. Rules are the floor, not the ceiling.
FOR AI AGENTSEvery agent and every process has a named human who is accountable.
AI agents · Compliance · Integrity
Build the cockpit before you launch the AI agents.
AI agents act faster than any approval round. To keep them within their guardrails, you need what aviation has perfected over eighty years: a human in charge, a shared picture of the situation and rules that run in the system, not in a PDF.
PRIMARY FLIGHT DISPLAY · INTEGRITYLIVE
Pilot in commandHuman
Active agents3 · one picture
Guardrailsas code · active
Audit trailcomplete
74% · 21%plan to deploy AI agents · have mature governance (Deloitte 2026)
Dec 2027EU AI Act high-risk obligations, after the Digital Omnibus
22 airline captainsexaminers for the aviation authorities, on the Capitum team
90 daysfrom first picture to the first provably compliant lighthouse
D
Almost everyone uses AI. Hardly anyone can prove it operates with integrity.
Three show-of-hands questions: this is how audiences answer in our keynotes, industry after industry.
1Who already uses current AI models?Almost all
2Who achieves real efficiency gains?Some
3Who can prove compliance and integrity?Hardly anyone
Typical audience result, not a representative survey.
74% · 21%
74% of companies plan to deploy AI agents within two years. Only 21% have a mature governance model for them.
Source: Deloitte, State of AI in the Enterprise, 2026Dec 2027
The EU AI Act high-risk obligations were postponed. The question of whether you can evidence your AI processes was not.
Digital Omnibus, agreement of Parliament and Council, June 2026C
Four cockpit rules apply one to one to AI agents.
In the cockpit, rules protect lives. Crews make the right call under pressure because they behave in learnable ways and because everything around them makes that behaviour easy.
The co-pilot calls “go-around”
Whoever sees something speaks up early, even to the captain.
FOR AI AGENTSThe agent speaks up before the error happens, not in the audit.
Never trust the autopilot blindly
Danger starts when everything looks calm.
FOR AI AGENTSChecking is a matter of design, not discipline.
Deciding under pressure
First the picture, then the options, then the decision. Even when time is short.
FOR AI AGENTSAgents deliver facts and options. The human decides the edge cases.
Trust in AI is not a feeling. It is a system property. You can build it.
The model
Integrity belongs at the centre, where human judgement and machine capability meet.
The behaviours that keep an aircraft out of trouble are the same ones that keep an organisation honest. AI agents can carry that integrity into daily work instead of parking it in a policy nobody opens.
- 1Human and machine as one crew. Clear roles, clear handovers.
- 2High-performance teams. Psychological safety, so mistakes surface early.
- 3Resilience. When conditions turn, the system holds.
- 4Scaling integrity. Beyond individual heroes, into every process.
W
From chatbot to autonomous enterprise.
Execution shifts from manual queries to collaborative agent systems. With every wave the value rises, and with it the duty to make boundaries visible and verifiable.
WAVE01
Chatbots
Assisted interaction
HUMAN EXECUTINGWAVE02
AI-augmented
Automated workflows
HUMAN IN THE LOOPWAVE03
AI agents
Autonomous execution
AGENT EXECUTINGWAVE04
Agentic enterprise
Multi-agent ecosystems on certified data
HUMAN-LED STRATEGIC OVERSIGHTChatbots · CopilotsAI-powered workflowsAutonomous unitsMulti-agent systems →
S
The best compliance nudge arrives in the second the mistake is made.
Integrity grows in three stages. Try the compliance co-pilot below for yourself.
STAGE 1 · TODAY
The AI co-pilot
Nudges right in the workflow, before the mistake happens.
Human decides every caseSTAGE 2 · 2026
Guardrails as code
Certified data products, rules run in the system.
Human defines the rulesSTAGE 3 · 2027
Multi-agent systems
Agents check each case, live and with an audit trail.
Human is pilot in commandSkip the second stage and you will fail at the third.
LIVE-DEMO
A
Without certified data, the autonomous enterprise remains a pilot project.
Almost every company has AI pilots; few scale them. The bottleneck is rarely the model. It is data that is scattered, uncertified and legally unclear. Integrity cannot be written into a 150-page PDF. It belongs in the code.
DataThe content itself
+
CodeHow it is produced and refreshed
+
MetadataMeaning, lineage, quality
+
PolicyWho may use it for what
+
ContractService level and schema promise
=
Data productCertified, with an owner and a guarantee, ready to use
| Raw data | Certified data product | |
|---|---|---|
| Accountability | Nobody, or IT | A named owner in the business |
| Quality | Unknown until something breaks | Measured, with a service level |
| Legal certainty | Checked per request | Purpose and consent in the code |
| Access | Ticket, weeks | Self-service, minutes |
| Reuse | Copied, diverging versions | One source, many users |
Computational governance: five automatic checks on every data access
Before: request, ticket, committee, manual review, approval. Often weeks, never repeated.
Data contract
Schema, quality and service level agreed and versioned.
Schema check
Every change is checked before release.
Purpose limitation
Data usable only for approved purposes.
Pseudonymisation
Personal data masked continuously.
Audit trail
Every access logged tamper-proof.
Humans set the rules and decide the exceptions. Machines enforce them, every time.
TIER 3
Business automation and AI agents
Agents, decision services, workflows, straight-through processing
TIER 2
Governance layer as code
Policy as code, data contracts, active metadata graph, access and masking
TIER 1
Federated business domains
Each domain builds and owns its data products, for example Medical, Commercial, Supply, Finance
revFADP · GDPREU AI ActEU Data ActFINMA-RS 2023/1GxP · EU-GMP Annex 11
B
Agents check each case. The human remains pilot in command.
Prototype at a bank in Spain: a Design Thinking for AI workshop produced an ecosystem of three specialised agents for SME credit assessment. Each agent has a role, a handover and an audit trail.
HUMAN
Request
Submits the case, for example a loan application from Madrid.
AGENT 1
Extraction
Reads tax and financial documents and pulls out the facts.
AGENT 2
Compliance & risk
Checks live via API against registers and credit bureaus.
AGENT 3
Preparation
Drafts the decision memo with its reasoning.
HUMAN
Approval
Decides the edge cases and signs off.
9.6 secinstead of 3 daysPre-assessment lead time in the prototype
E
Stop counting agents. Count pictures.
From the essay “Build the Cockpit Before You Add the Agents” by Michael Lewrick.
An air traffic controller handles up to eighteen aircraft at once. A capable engineer loses track of five AI agents within two hours. The difference is not talent. It is structure.
- 1948 · MackworthIn pure monitoring tasks, detection drops by 10 to 15 percent within the first half hour.
- 1973 · Chase & SimonChess masters do not have bigger memories but bigger units. Expertise makes the chunks larger.
- 1983 · Bainbridge“Ironies of Automation”: the more reliable the system, the more the skill for the rare emergency withers.
- 2010 · Parasuraman & ManzeyAutomation complacency does not disappear with practice. Only the design of the task helps.
Four moves for this quarter, ranked by leverage:
1
Count pictures, not agents
How many independent initiatives does one person touch per day? Anything above two is your real overload.
2
Build the status field
One file per initiative, filled in by every agent at every stop. Readable in thirty seconds.
state: contract under review · last_decision: routing via CH approved · open_question: clause 48 confirmed? · next_step: legal sign-off
3
Make the handover an act
Read the status field, resolve open questions, then take over explicitly, like two controllers at shift change.
4
Named review, no more than 30 minutes at a time
Whoever reviews agent output for an hour no longer reviews the second half. They sign it.
9
Don’t wait for perfect regulation. Build guardrail-ready today.
The next 90 days: from first picture to the first provably compliant lighthouse.
WEEKS 1–4
Make it visible
- Map AI usage, including shadow AI
- Map value streams and data
- Pick one high-risk workflow
MILESTONETop 2 candidates and business case
WEEKS 5–8
Co-pilot and data
- Switch on nudges in the workflow
- First certified data product
- Data contract and policies as code
MILESTONEGovernance and data quality live
WEEKS 9–12
Make it provable
- Audit trail active
- Launch an agentic workflow
- Measure against the baseline
MILESTONELighthouse live, executive scorecard
F
From the cockpit to the boardroom. And into the code.
Leadership and technology from one source: Capitum brings the crew principles, Lewrick & Company the implementation with AI agents and data.
KEYNOTE · LEWRICK & COMPANY AND CAPITUM
The Future of Integrity: Where cockpit excellence meets agentic AI
Two perspectives on one stage: what crews get right under pressure, and how AI agents carry that integrity into daily work.
Request the keynotePART ICockpit & compliance · Dr. Martin Weth
PART IIImplementation & agentic AI · Dr. Michael Lewrick
CLOSEQuestions, discussion, drinks
LEWRICK & COMPANY1 DAY · C-LEVEL
Executive Discovery Workshop
- Shared ambition for automation
- Top use cases and risk appetite
- Maturity heatmap
LEWRICK & COMPANY3–4 WEEKS · CDO, IT, LEGAL
Regulatory Data Architecture Review
- Gap analysis against revFADP, GDPR, AI Act and sector rules
- Target architecture
- Prioritised roadmap
LEWRICK & COMPANY90 DAYS · ONE DOMAIN
Enterprise Automation Pilot
- Lighthouse data product
- Policies as code in production
- Live automation with scorecard
CAPITUMWORKSHOP
Cockpit excellence for leadership teams
Decision-making, compliance & integrity and high-performance culture, taught by airline captains.
CAPITUMSIMULATION LAB
Deciding under pressure, experienced
Leadership teams train crew principles in realistic scenarios.
JOINTTAILORED
Integrity by Design
Leadership and technology in one programme: crew principles plus guardrails as code.
K
The crew.
From the cockpit. Into the boardroom. Into the code.
Dr. Martin Weth
Managing Partner · Capitum Consulting
25 years of experience in flight operations at international airlines and an internationally recognised executive trainer.
Capitum has its roots in technical aviation consulting. With a team of 22 airline captains, all examiners on behalf of the civil aviation authorities, Capitum is the leading provider for transferring leadership principles from the cockpit to the business world.
Dr. Michael Lewrick
Managing Director · Lewrick & Company
Expert in AI and innovation, author of “AI and Innovation” and the Design Thinking series (Wiley), Thinkers50 nominee in 2022 and 2023.
Guides regulated companies from AI pilots to scaled, provably compliant AI portfolios, with AI agents in the workflow and human-in-the-loop design.
V
The Future of Integrity in Vienna.
Keynote and exchange with Dr. Martin Weth and Dr. Michael Lewrick, November 2026. Photos from the evening will follow.
Contact
AI delivers energy and intelligence. Integrity is ours to build.
Those who lay the guardrails today will fly faster and safer tomorrow. In a first 60-minute conversation we work out which entry point fits your organisation.