🎮 The Next Input — Issue #154

In partnership with

⚡ The Briefing — 60 sec

• As more Americans adopt AI tools, fewer say they can trust the results Not so subtle shoutout to my own business Cylentis.com. More people are using AI, fewer people trust it, and that gap right there is basically the whole game now.

• AI is being blamed for job losses and those using it are warning this may just be the start Adapt or lose your job? That is pretty much where this is headed. The ABC piece even quotes a former PwC employee saying, “adopt AI or die,” which is about as subtle as a brick through a window.

• Wikipedia bans AI-generated articles Bout time. Wikipedia’s English edition now prohibits editors from using generative AI to generate or rewrite articles, with only limited exceptions, which feels like one of the clearest “enough of the slop” moments yet.

🛠️ The Playbook — The Trust Layer Engine

Mission
Build AI workflows that are actually useful and actually trusted, instead of fast, shiny, and quietly dubious.

Difficulty
Intermediate

Build time
3–5 hours

ROI
Higher adoption, better outputs, and a much stronger case for AI systems that people will rely on instead of side-eye.

0) Why This Matters

This is the split defining AI right now.

Usage is up, trust is down. TechCrunch cites a Quinnipiac poll showing only 21% of Americans trust AI-generated information most or almost all of the time, while 76% trust it only rarely or sometimes; at the same time, the share who say they have never used AI tools fell to 27%, down from 33% in April 2025.

Then there is the labour angle. The ABC piece quotes a former PwC employee who helped build autonomous agents saying “adopt AI or die,” while describing efforts to automate routine back-office work and even build multi-agent systems for clients.

And then Wikipedia stepped in and basically said: enough. According to Information Age, the English Wikipedia now prohibits editors from using generative AI tools to generate or rewrite articles, because LLM output often conflicts with core content policies around neutrality, verifiability, and no original research.

So the play is not just “use AI more.”

It is:

• build workflows people can verify

• keep provenance attached

• use AI where it speeds up work, not where it muddies truth

• treat trust as part of the product, not a bonus feature

1) Architecture

2) Workflow

1. Pick one recurring workflow where people already want faster answers but still need confidence in the result.

2. Connect the workflow to trusted source material instead of letting the model improvise from thin air.

3. Require the model to produce answers with citations, evidence snippets, or source references.

4. Route higher-stakes outputs through human review before they are sent or acted on.

5. Track where users accept the answer, edit the answer, or ignore the answer completely.

6. Improve the workflow by fixing retrieval, prompts, or review rules based on where trust breaks.

3) Example Prompts

Grounded Answer Prompt

You are answering a question using only the supplied source material.

Rules:
- do not invent facts
- cite the relevant source snippets
- if the evidence is weak or missing, say so clearly
- keep the answer concise and operational

Question:
[insert question]

Sources:
[insert sources]

Trust Review Prompt

You are reviewing an AI output for trustworthiness.

Check:
- whether every key claim is supported
- whether the tone sounds more certain than the evidence allows
- whether any unsupported inference slipped in
- whether the answer should be accepted, edited, or rejected

Return:
1. decision
2. reason
3. corrected version if needed

Workflow Fit Prompt

You are assessing whether a workflow is suitable for AI assistance.

For the process below, identify:
- which steps can be sped up safely
- which steps require citations or provenance
- which steps must stay human
- the top 5 trust risks

Process:
[insert workflow]

Adoption Gap Prompt

You are diagnosing why users are not trusting an AI workflow.

Given the workflow, outputs, and user feedback:
- identify whether the issue is retrieval, hallucination, tone, or lack of evidence
- identify the biggest trust-breaking pattern
- recommend one concrete fix

Return 3 bullet points only.

4) Guardrails

• No citation, no answer.

• Do not let the model rewrite truth-heavy content without source grounding.

• Separate speed gains from trust gains.

• Require human review for anything legal, financial, board-facing, or reputational.

• Track where users ignore AI outputs, not just where they open them.

• Treat slop as a systems problem, not just a prompt problem.

5) Pilot Rollout — 3 hours

1. Pick one workflow where trust matters more than novelty.

2. Connect one clean source system and one model to that workflow.

3. Force the output to include source references or evidence citations.

4. Add a simple human review step for sensitive cases.

5. Run 10–20 live examples and record acceptance, correction, and rejection rates.

6. Tighten retrieval and prompts before rolling the workflow wider.

6) Metrics

• Answer acceptance rate

• Human correction rate

• Percentage of outputs with valid citations

• User trust score

• Time saved per workflow

• Number of unsupported claims detected

• Percentage of users returning to the workflow

Pro Tip: Adoption without trust is just curiosity with a countdown timer on it.

🎯 The Arsenal — Tools & Platforms

• Azure AI Search · retrieval layer for grounding answers in real enterprise content · Azure AI Search

• Pinecone · semantic retrieval for evidence-backed workflows · Pinecone

• Claude / GPT / Gemini · useful model layer, but only when tied to proof instead of vibes · Anthropic · OpenAI · Gemini

• Google Sheets / Airtable · lightweight trust dashboard for corrections, approvals, and adoption · Google Sheets · Airtable

• Wikipedia’s new policy · probably the clearest public reminder this year that “AI-generated” is not the same thing as “good enough” for serious knowledge work.

Copy-paste prompt block:

You are helping me design a Trust Layer Engine for an AI workflow.

For the workflow below:
1. identify the trusted source systems
2. identify where AI can help safely
3. identify where citations or provenance are mandatory
4. identify which steps must stay human
5. identify the top 5 trust risks
6. design a simple review process
7. propose a 2-week pilot

Workflow:
[insert workflow here]

Return the answer in markdown with sections for:
- Workflow summary
- Source systems
- AI-assisted steps
- Human-only steps
- Trust risks
- Review process
- Pilot rollout
- Metrics

💡 Free Office Hours

If you are trying to build AI workflows people will actually trust instead of merely tolerate, I run free office hours to help map the workflow, the provenance layer, and the fastest pilot path.

Book here: https://calendly.com

88% resolved. 22% stayed loyal. What went wrong?

That's the AI paradox hiding in your CX stack. Tickets close. Customers leave. And most teams don't see it coming because they're measuring the wrong things.

Efficiency metrics look great on paper. Handle time down. Containment rate up. But customer loyalty? That's a different story — and it's one your current dashboards probably aren't telling you.

Gladly's 2026 Customer Expectations Report surveyed thousands of real consumers to find out exactly where AI-powered service breaks trust, and what separates the platforms that drive retention from the ones that quietly erode it.

If you're architecting the CX stack, this is the data you need to build it right. Not just fast. Not just cheap. Built to last.

See the data

🕹️ Game Over

Everyone wants faster AI. The winners will be the ones who can prove it is worth trusting.

— Aaron Automating the boring. Amplifying the brilliant.

Subscribe: link