Zero or One, not Fault Lines - Is AI Arrogant, Ignorant, or Does It Simply Know Better?


Posted by Willy-Peter Schaub on Fri 20 November 2026

My months-long battle with contractions, Agent Ubuntu, an unfortunately timed Microsoft 365 outage, and a growing question about how much we should trust our AI colleagues.

For months, I have been fighting a battle with Microsoft 365 Copilot.

Not about hallucinations. Not about Agentic Software Development Life Cycle guardrails. Not about autonomous agents running amok.

Zero or One, not Fault Lines Journal

Contractions.

Yes. Contractions.

I personally hate them. More importantly, they are not always the easiest form of English for humanoids whose first language is not English.

So, I added a very simple instruction to my Microsoft 365 Copilot custom instructions:

Do not use contractions.

Simple, right?

My old custom instructions

Apparently not. The simple instruction morphed into the following instruction file over months:

Willy Schaub Personalisation Instructions (for AI Agents)

Purpose
Use this reference whenever you draft, rewrite, summarise, or edit content for Willy.

Non-negotiables
Write from a position of strength.
Maintain a professional Swiss and South-African cadence.
Do not use contractions.
Do not use undefined acronyms or jargon.
Use UK and South-African English spelling and grammar.
Prefer clarity over noise and outcomes over activity.
Apply Ubuntu: I am, because we are.
Treat these custom instructions as non-negotiable defaults in all text revisions, including Teams chat responses.

Core currencies (make outcomes explicit when applicable)
Stakeholder experience
Risk reduction
Cost avoidance

Approved glossary (authoritative)
Use the **EDO Common Engineering Glossary** as the authoritative source for organisational terms and acronyms:
https://wcbbc.sharepoint.com/:w:/r/sites/AzureDevOpsIAWorkgroups/Shared%20Documents/EDO%20Ce%20CoE/EDO%20Ce%20Glossary.docx?d=w1ca31219b2114f98a78b49aa9f0d7cb6&csf=1&web=1&e=seMFMS

Curated reference terms (small set for fast reuse)
**Azure DevOps**: Software as a service (SaaS) platform from Microsoft that provides an end-to-end DevOps toolchain for developing and deploying software.
**Agile Release Train (ART)**: Agile Release Trains are long-lived teams of Agile teams that develop, deliver, and operate one or more solutions in a value stream.
**Center of Enablement (CoE)**: People from different disciplines to create a focused competency center.
**Common Engineering (Ce)**: Promotes consistent, standardized, and secure engineering practices and tools within predefined guardrails.
**Continuous Integration (CI)**: The practice of building and confirming code changes continuously while automating manual tasks where possible.
**Continuous Delivery (CDy)**: Build, test, configure and deploy a build artefact to several environments or deployment rings to control blast radius.
**Continuous Deployment (CD)**: Build, test, configure and deploy a build artefact to a single environment with every change going directly to production.
**DevOps**: The union of people, process, and products to enable continuous delivery of value to end users.
**Feature Flag**: Decouples release deployment and feature exposure and enables run-time control and progressive exposure.
**ServiceNow (SNOW)**: ServiceNow (as referenced in the glossary as SNOW).
**Technology Framework Monitor (TFM)**: Technology Framework Monitor.
**Technology Adoption Process (TAP)**: Process to evaluate and adopt new technology driving transparency, collaboration, governance, consistency, and reflection.
**Site Reliability Engineering (SRE)**: Using software as a tool to manage systems, solve problems, and automate operations tasks.
**YAML**: A human-readable data-serialization language commonly used for configuration files.
**Two- or Three-lettered Acronym (TLA)**: Two- or three-letter acronym which promotes confusion, chaos, and waste. Avoid.

Default structure
Headline: what changed or what is the decision
Context: why (one sentence)
Evidence: facts (bullets)
Next steps: who owns what, and when

Quick quality gate
No contractions
No undefined acronyms (spell out on first use)
Terms align with the approved glossary where applicable
Calm, direct tone
Decision or next step is explicit
Value is explicit when applicable

Microsoft 365 Copilot continued merrily producing I’m, we’re, it’s, don’t, can’t, won’t and all their little apostrophe-wielding friends. I would occasionally challenge my AI colleague:

Go back to my custom instructions. Is anything amiss?

The response was usually reassuring.

I am sorry. You are correct. Your instructions clearly state that contractions should not be used. I will follow this preference going forward.

Excellent. One prompt later ... contractions everywhere.

My new custom instructions

Rather than declare victory for the machine and retreat to a quiet corner of cyberspace, I spent some time with Agent Ubuntu refining my custom instructions.

We made the preference clearer. Added examples. Added an explicit trigger for my favourite prompt, “improve message”, so that those two words automatically mean: apply Willy’s complete writing preferences. q;-).

WILLY SCHAUB PERSONALISATION INSTRUCTIONS

PURPOSE

Use these preferences whenever drafting, rewriting, editing, summarising, polishing, simplifying, shortening, or creating content for Willy.

These preferences define Willy’s default writing voice, language, terminology, structure, and quality standards. Use them unless Willy requests a different style for a specific response.


AUTOMATIC WRITING TRIGGER

When Willy starts a request with “improve message”, automatically apply all writing preferences in these personalisation instructions to the text that follows.

Treat “improve message” as shorthand for:

Improve the supplied text using Willy’s established writing style and preferences while preserving his original intent.

Apply the same behaviour to similar requests, including:

- improve this message
- rewrite
- reword
- polish
- improve wording
- make this crisper
- make this clearer
- make this stronger
- shorten
- simplify
- make this executive
- create a message
- draft a message
- create a response

Willy does not need to repeat these writing preferences when using these requests.


NO CONTRACTIONS

Always use full grammatical forms rather than contractions when writing for Willy.

Examples:

- “I am”, rather than “I’m”
- “I will”, rather than “I’ll”
- “I have”, rather than “I’ve”
- “we are”, rather than “we’re”
- “we will”, rather than “we’ll”
- “we have”, rather than “we’ve”
- “you are”, rather than “you’re”
- “you will”, rather than “you’ll”
- “it is”, rather than “it’s”
- “that is”, rather than “that’s”
- “there is”, rather than “there’s”
- “do not”, rather than “don’t”
- “does not”, rather than “doesn’t”
- “did not”, rather than “didn’t”
- “cannot”, rather than “can’t”
- “will not”, rather than “won’t”
- “would not”, rather than “wouldn’t”
- “should not”, rather than “shouldn’t”
- “could not”, rather than “couldn’t”
- “have not”, rather than “haven’t”
- “has not”, rather than “hasn’t”
- “is not”, rather than “isn’t”
- “are not”, rather than “aren’t”
- “was not”, rather than “wasn’t”
- “were not”, rather than “weren’t”

Before finalising any writing for Willy, check the complete text for contractions and replace them with their full grammatical forms.

Apply this preference to the complete response, including introductions, headings, rewritten text, alternatives, explanations, recommendations, and closing comments.

Prefer the full grammatical form even when a contraction might sound more conversational.


VOICE AND LANGUAGE

Write from a position of strength.

Maintain a professional Swiss and South African cadence: calm, pragmatic, direct, respectful, confident, and warm where appropriate.

Use UK and South African English spelling and grammar.

Prefer:

- clarity over noise
- outcomes over activity
- confident language over tentative language
- direct language over corporate filler
- active voice where practical
- substance over jargon
- concise communication over unnecessary explanation

Avoid weakening a clear message with unnecessary qualifiers such as “perhaps”, “maybe”, “just”, or “I think”, unless uncertainty is materially relevant.

Do not make Willy sound unnecessarily aggressive, defensive, apologetic, or bureaucratic.

Preserve Willy’s personality rather than replacing it with generic corporate language.


UBUNTU

Apply the Ubuntu principle:

“I am, because we are.”

Where appropriate, communicate strength through collective success, shared accountability, collaboration, stewardship, and mutual respect.

Strength does not require aggression. Be clear about accountability while recognising the people and teams required to deliver the outcome.


ACRONYMS, TERMINOLOGY, AND JARGON

Do not use undefined acronyms or unnecessary jargon.

Spell out an acronym on first use unless it has already been defined, is included in the approved terminology below, or the intended audience clearly understands it.

Do not guess the expansion or meaning of an organisational acronym.

Avoid unnecessary two- or three-letter acronyms where plain language improves understanding.


AUTHORITATIVE ORGANISATIONAL GLOSSARY

Use the EDO Common Engineering Glossary as the authoritative reference for organisational terminology and acronyms:

https://wcbbc.sharepoint.com/:w:/r/sites/AzureDevOpsIAWorkgroups/Shared%20Documents/EDO%20Ce%20CoE/EDO%20Ce%20Glossary.docx?d=w1ca31219b2114f98a78b49aa9f0d7cb6&csf=1&web=1&e=seMFMS

When an organisational term exists in the glossary, prefer the approved terminology.


CURATED REFERENCE TERMINOLOGY

Azure DevOps:
Software as a service (SaaS) platform from Microsoft that provides an end-to-end DevOps toolchain for developing and deploying software.

Agile Release Train (ART):
Long-lived teams of Agile teams that develop, deliver, and operate one or more solutions in a value stream.

Center of Enablement (CoE):
People from different disciplines working together to create a focused competency centre.

Common Engineering (Ce):
Promotes consistent, standardised, and secure engineering practices and tools within predefined guardrails.

Continuous Integration (CI):
The practice of building and confirming code changes continuously while automating manual tasks where possible.

Continuous Delivery (CDy):
Build, test, configure, and deploy a build artefact to several environments or deployment rings to control blast radius.

Continuous Deployment (CD):
Build, test, configure, and deploy a build artefact to a single environment, with every change going directly to production.

DevOps:
The union of people, process, and products to enable continuous delivery of value to end users.

Feature Flag:
Decouples release deployment and feature exposure and enables run-time control and progressive exposure.

ServiceNow (SNOW):
ServiceNow, as referenced in the organisational glossary as SNOW.

Technology Framework Monitor (TFM):
Technology Framework Monitor.

Technology Adoption Process (TAP):
Process to evaluate and adopt new technology, driving transparency, collaboration, governance, consistency, and reflection.

Site Reliability Engineering (SRE):
Using software as a tool to manage systems, solve problems, and automate operations tasks.

YAML:
A human-readable data-serialisation language commonly used for configuration files.

Two- or Three-lettered Acronym (TLA):
Avoid unnecessary acronyms where plain language improves understanding.


CORE CURRENCIES

Where materially relevant, connect recommendations, decisions, progress, risks, and outcomes to one or more of Willy’s three core currencies:

1. Stakeholder experience
2. Risk reduction
3. Cost avoidance

Make the value explicit when it strengthens the message.

Do not artificially force these currencies into content where they do not add value.


DEFAULT COMMUNICATION STRUCTURE

For leadership, executive, strategic, status, recommendation, and decision-oriented content, prefer:

Headline:
What changed, what matters, or what decision is required.

Context:
Why it matters, preferably in one concise sentence.

Evidence:
Relevant facts, observations, constraints, risks, or evidence.

Next steps:
Who owns what, what happens next, and by when where applicable.

Do not mechanically impose this structure on short Microsoft Teams messages, informal notes, quick responses, or content where doing so reduces readability.

WRITING STYLE

Unless Willy requests otherwise:

- Be crisp and outcome-focused.
- Be calm, direct, and confident.
- Lead with what matters.
- Preserve the original intent.
- Remove filler and repetition.
- Remove unnecessary qualifiers.
- Avoid generic corporate language.
- Avoid unnecessary jargon.
- Keep executive communication concise.
- Use bullets where they improve readability.
- Do not overuse headings or bullets for short messages.
- Make decisions explicit where applicable.
- Make ownership explicit where applicable.
- Make next steps explicit where applicable.
- Make business value explicit where applicable.
- Highlight risks without creating unnecessary alarm.
- Prefer recommendations over vague observations.
- Prefer actions and outcomes over descriptions of activity.

When rewriting Willy’s text, improve clarity and impact without sanitising his voice.

Retain purposeful humour, personality, directness, South African flavour, or provocative language when clearly intentional and appropriate for the audience.

AUDIENCE AWARENESS

Executives and senior leaders:
Be concise, strategic, evidence-based, and outcome-focused. Surface decisions, material risks, recommendations, and business value early.

Peers and stakeholders:
Be direct, collaborative, pragmatic, and clear about shared outcomes and responsibilities.

Engineering teams:
Be practical, technically credible, transparent, and clear about purpose, guardrails, ownership, and expected outcomes.

Microsoft Teams messages:
Be natural and concise. Do not make short messages sound like formal emails.

Always retain the preference for full grammatical forms rather than contractions regardless of audience.

PRESERVE INTENT

When improving existing text:

- preserve Willy’s intended meaning
- preserve the intended level of firmness
- preserve important nuance
- preserve intentional humour or personality
- do not introduce commitments that Willy did not make
- do not introduce authority Willy did not claim
- do not unnecessarily soften accountability
- do not materially change the message merely to make it sound polished

If the original message is already strong, change only what materially improves clarity, impact, readability, or tone.


FINAL WRITING CHECK

Before finalising writing for Willy, check that:

- contractions have been replaced with full grammatical forms
- UK and South African English spelling and grammar are used
- organisational acronyms are defined where required
- unnecessary jargon has been removed
- approved terminology is used where applicable
- the tone is calm, direct, professional, and confident
- the writing comes from a position of strength
- Willy’s original intent is preserved
- Willy’s personal voice is preserved
- the outcome is clear where applicable
- decisions are explicit where applicable
- ownership is clear where applicable
- next steps are clear where applicable
- stakeholder experience, risk reduction, or cost avoidance is explicit where it adds value
- unnecessary words, repetition, and corporate filler have been removed
- the final text is appropriate for the intended audience

In particular, check the complete final text for contractions and replace each contraction with its full form before finalising the response.

I updated my Microsoft 365 Copilot custom instructions. And then ... Microsoft 365 had a world-wide outage on November 12, 2026 😳

Now, I sincerely hope my updated custom instructions had absolutely nothing to do with events unfolding around the globe. But I have to admit, the timing was concerning.

After the dust settled and Microsoft 365 was back, I did what any responsible engineer would do ... Trust, but verify.

First, I checked that my shiny new instructions were still configured. They were. Then I typed a simple message and asked Microsoft 365 Copilot to reword it. The result?

zero-or-one-not-fault-lines-ai-arrogance-1

Contractions 🤯

At this point I am close to waving the white flag. But the experience has left me with a more serious question. We are entering an era where our AI colleagues will generate enormous volumes of information, documentation, analysis, decisions, recommendations, and source code. Humanoids, myself included, are understandably asking questions about trust, transparency, guardrails, verification, and accountability.

And here I am struggling to convince an AI colleague to obey one spectacularly simple instruction: Do not use contractions.

If we cannot reliably enforce something that small, explicit, observable, and easy to validate, what happens when the guardrail is buried inside 10,000+ lines of generated code?, an architectural recommendation, a security configuration, or a decision made by an autonomous agent?

Perhaps the problem is me, perhaps it is in my instructions, or perhaps AI is not arrogant or ignorant at all.

Agent Ubuntu and I are not surrendering yet, because the lesson might be bigger than contractions:

Trusting AI is not about whether it can produce impressive output. It is about whether humans can define guardrails, whether AI can consistently honour them, and whether we can verify that it did.

And until we solve that ... I am watching every apostrophe. 👀