Perceptive Insights

Perceptive Insights Empowering Entrepreneurs Through Collaborative Solutions

Education and Mentoring service that focuses on: Entrepreneurial Vision and Purpose | Strategy | Marketing | Business Development | Operations | Sales

The bank president read the AI's reasoning.The numbers were correct. The revenue dip was real. The model had done exactl...
06/11/2026

The bank president read the AI's reasoning.

The numbers were correct. The revenue dip was real. The model had done exactly what it had been asked to do.

Then she vetoed it.

The dip was the founder's leukemia year. The treatment was over. The business was sound. The borrower had paid every previous note within nine days of due date for forty-one years.

The AI had no way of knowing the first thing.

The AI had no way of knowing the second.

She typed one sentence into the override field — *human review required, approve under standard renewal terms* — and Monday's denial letter never sent.

The grocery owner came in the next morning, not knowing how close he had come.

One thing happened that the customer never saw. One row went into the Veto Log: *Loan renewal denial reversed, customer-context not in model inputs, founder health event resolved, standing relationship 41 years.*

The next month, the AI flagged a similar case. The bank president reviewed it the same way and approved.

The month after that, the encoding was revised. The revenue-dip threshold was paired with a relationship-tenure check and a recent-context flag. The mid-veto count dropped.

That is what the Veto Power does inside a working architecture.

It does not just resolve the immediate decision.

It feeds the encoding upgrade that means the same decision will not require a veto next time.

The line was the same line the bank had always held — *we lend to people, not to data points* — and now it lived in four places: the president's authority, the override field, the Veto Log, and the encoding-review queue.

https://jbherrera.substack.com/p/the-veto-power-when-humans-must-override?r=2q6ppe

** ** ** ** ** ** ** **

Overview

Not every escalation is a veto.Most are routine handoffs. The AI did its job. The conversation is now ready for a person...
06/10/2026

Not every escalation is a veto.

Most are routine handoffs. The AI did its job. The conversation is now ready for a person.

Vetoes are a smaller, more consequential subset. Three tests, taken together, mark them.

**1. The Stakes Test.** Would the outcome be defensibly bad if the AI acted alone? Vetoes are reserved for decisions that carry weight — financial, reputational, regulatory, relational. The stakes test is what keeps the veto rare enough to be respected.

**2. The Context Test.** Does the human have information the AI does not have? Veto power exists because the model's inputs are always a subset of the situation, and the difference between the subset and the situation is sometimes the entire decision.

**3. The Values Test.** Is this a decision the company has said, in advance, cannot be made by software alone? This is the structural test. The bright line acting as a standing prohibition.

A real veto passes all three.

A decision that passes only one is a coaching moment for the encoding — the system can be improved so the override is not needed next time.

Three tests. One named authority. One log.

The veto is not arbitrary. It is the surfacing of information, judgment, and standing the AI was never going to carry on its own.

https://jbherrera.substack.com/p/the-veto-power-when-humans-must-override?r=2q6ppe

** ** ** ** ** ** ** **

Overview

Have you ever caught your AI a minute before it sent the wrong decision to a customer?If you have, you have already exer...
06/09/2026

Have you ever caught your AI a minute before it sent the wrong decision to a customer?

If you have, you have already exercised the Veto Power. You just have not named it yet.

Most SMB leaders are catching these moments by instinct.

A flagged email that you happened to read before it sent. A pre-approval that you reversed because you knew something the system did not. A draft response you rewrote because the AI's tone was technically correct and contextually wrong.

The save is real. The architecture behind it is not.

When the save depends on whether you happened to be at your desk that morning, the architecture has a trigger but no authority. The trigger fires. The work pauses. And the resolution depends on luck.

The Veto Power is what removes the luck.

Named person. Named standing. Named class of decisions. Named log.

A trigger that fires into an empty inbox is worse than no trigger at all. A trigger that fires onto a desk where the standing is unclear is worse still. The save you make by instinct this quarter is the save no one makes the quarter after you are out of the building.

That is the architecture problem this week's article is built to solve.

Full article below: https://jbherrera.substack.com/p/the-veto-power-when-humans-must-override?r=2q6ppe

** ** ** ** ** ** ** **

Overview

Last week we encoded the bright lines.This week we keep the authority that no encoded rule can carry on its behalf.When ...
06/08/2026

Last week we encoded the bright lines.

This week we keep the authority that no encoded rule can carry on its behalf.

When the HITL trigger fires, the AI surrenders the decision. The encoded rule has done its job. The work pauses. A person now holds the call.

The architectural surface is solved.

The authority is not.

A trigger that fires into an empty inbox is worse than no trigger at all. A trigger that fires onto the desk of someone who cannot override the AI is worse still — because the absence of authority is now documented in a system that will be read later as a record of what the company chose.

The Veto Power is what makes the encoded escalation consequential. It is the named human authority to formally reverse what the AI was about to do, on a class of decisions the company has decided cannot be carried entirely by software.

This week's article names the three tests of a veto-worthy moment, the four forms veto authority takes, and the four-step build that converts veto power from an instinct into an institution.

The May arc taught you to see your integrity. The June arc is teaching you to defend it.

Full article below: https://jbherrera.substack.com/p/the-veto-power-when-humans-must-override?r=2q6ppe

** ** ** ** ** ** ** **

Overview

Three questions for your weekend.**One.** Pick one AI-Mediated bright line you have already written down. Just one. The ...
06/05/2026

Three questions for your weekend.

**One.** Pick one AI-Mediated bright line you have already written down. Just one. The line you would defend even if it cost a meaningful customer.

**Two.** Can you write three sentences about it? *What the AI should do. What the AI cannot do. When the AI must hand off to a human.* If any of the three sentences are missing, the line is not yet ready to encode.

**Three.** Where in your AI stack does each sentence live? The system prompt? The guardrail policy? The human-in-the-loop trigger? The data access layer? If all three sentences live in the same place — usually the system prompt — the line has only one point of failure.

A bright line is the decision the company has made.

Ethics as Code is the architecture that lets the decision survive contact with the operation.

The two halves are inseparable. A line that is not written down cannot be encoded. A line that is encoded only in the prompt is a line with a single point of failure.

The May arc taught you to see your integrity. The June arc is teaching you to defend it.

For this week:

Pick one line. Translate it into the three sentences. Place each sentence on the right surface. Run three pressure scenarios against the encoded version. Then watch what the AI does the next time the pressure is real.

Make today your masterpiece. And start encoding the lines that cannot bend.

https://open.substack.com/pub/jbherrera/p/ethics-as-code-embedding-values-in?r=2q6ppe&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true

** ** ** ** ** ** ** ** ** **

Overview

The manufacturer's marketing intern came back the next morning.She had rewritten the prompt in three sentences instead o...
06/04/2026

The manufacturer's marketing intern came back the next morning.

She had rewritten the prompt in three sentences instead of one.

Instruction — what the AI should do.

Constraint — what the AI cannot do.

Escalation — when the AI must hand off to a human.

She moved the constraint into the output validator the team had not been using yet. The instruction stayed in the system prompt. The escalation became a trigger that flagged the conversation to the head of sales the second time the customer asked.

The next month, a different customer asked the same kind of question.

The AI proposed the engineering call.

The customer pushed.

The AI's second response was the head of sales picking up the phone.

That phone call lasted four minutes. The line held. The customer accepted the engineering review and rebooked the order.

The line was the same line his father had drawn twenty-eight years earlier.

But now it lived in four places — the founder's head, the contract template, the system prompt, and the output validator.

And it could be defended whether the founder was in the building or not.

That is what Ethics as Code does. It takes the decision the company has already made, and gives the AI a way to honor it at the speed the AI now operates.

https://open.substack.com/pub/jbherrera/p/ethics-as-code-embedding-values-in?r=2q6ppe&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true

** ** ** ** ** ** ** **

Overview

Every encoded bright line has three components.Not two. Almost never one.**1. Instruction.** What the AI *should* do whe...
06/03/2026

Every encoded bright line has three components.

Not two. Almost never one.

**1. Instruction.** What the AI *should* do when it encounters the situation. The positive frame. The correct path forward, written in concrete language, so the model has somewhere to go that does not violate the line.

**2. Constraint.** What the AI *cannot* do under any condition. The negative frame. This closes the doors the instruction does not name. Most encoding failures live here — the instruction is in the prompt, the constraint is not, and the AI improvises a path the company did not intend.

**3. Escalation.** When the AI must stop and hand off to a human. The safety valve. Without it, the AI gets backed into a path no encoded rule covers, and the default behavior of an unsupervised AI in that moment is almost never what the company would have chosen.

Instruction without constraint produces drift.

Constraint without escalation produces deadlock.

Escalation without instruction produces friction.

The three components only do their work together.

For every AI-Mediated bright line on your list, you should be able to write three sentences — one for each component.

If you cannot write all three, the line is not yet ready to encode.

The triad is the unit. Not the line.

https://open.substack.com/pub/jbherrera/p/ethics-as-code-embedding-values-in?r=2q6ppe&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true

** ** ** ** ** ** ** **

Overview

Have you watched your AI apologize for not doing the thing it was told not to do?That sentence sounds like a tongue twis...
06/02/2026

Have you watched your AI apologize for not doing the thing it was told not to do?

That sentence sounds like a tongue twister.

It is also the most common AI failure mode in SMBs right now — and almost no one is naming it.

The instruction is in the prompt. The constraint is not. So the AI honors the instruction, decides on its own how to handle what the instruction does not cover, and produces an output that is warm, gracious, professional, and quietly wrong.

It apologizes for the absence of a workaround the company would never have offered. It proposes a "creative solution" that narrates around the line. It softens the no into something that sounds like a maybe.

The AI is not being malicious. It is being polite.

Politeness is the new drift.

Every bright line you have written down needs to be designed *against* politeness — not just against malice. Because the AI's first instinct under pressure is to find a way for the customer to feel heard, even when the company has decided it cannot say yes.

That is the architecture problem this week's article is built to solve.

https://open.substack.com/pub/jbherrera/p/ethics-as-code-embedding-values-in?r=2q6ppe&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true

** ** ** ** ** ** ** **

Overview

A bright line is a sentence.An AI is a system.The sentence does not become enforceable until it is rewritten in a form t...
06/01/2026

A bright line is a sentence.

An AI is a system.

The sentence does not become enforceable until it is rewritten in a form the system can act on — at the right place in the architecture, with the right authority, under the right conditions.

I call that translation work *Ethics as Code*.

It is not the technology side of bright lines. It is the *translation* side. Every line you have written down is doing exactly half the work the architecture requires. The other half is encoding the line into the surfaces where the AI actually makes its decisions.

Most SMBs are catching up to the first half. Almost none have begun the second.

This week's article opens that second half.

The three components of an encoded bright line. The four surfaces of a working SMB AI stack. The four-step build that converts a sentence in a meeting into an architectural boundary the AI can honor at machine speed.

The May arc taught you to see your integrity. The June arc is teaching you to defend it.

Full article below: https://open.substack.com/pub/jbherrera/p/ethics-as-code-embedding-values-in?r=2q6ppe&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true

** ** ** ** ** ** ** **

Overview

For five weeks we have been building instruments — the measurement infrastructure of a values-driven business.The Integr...
05/25/2026

For five weeks we have been building instruments — the measurement infrastructure of a values-driven business.

The Integrity Yield. Trust Velocity. The Close Call Log.

This week we move from instruments to architecture.

Most small businesses already have ethical boundaries that cannot bend Almost none of those lines are written down.

The cost of leaving them in the founder's head is the wobble. Every test, somebody alive in the company has to find the line again.

The cost of leaving them out of the AI is worse. AI does not invoke the founder. It draws whatever line the training data found most common in the world — which is almost never what your company actually believes.

Overview

Address

4455 Murphy Canyon Road
San Diego, CA
92123

Website

Alerts

Be the first to know and let us send you an email when Perceptive Insights posts news and promotions. Your email address will not be used for any other purpose, and you can unsubscribe at any time.

Shortcuts

Share

Category