🎮 The Next Input — Issue #150

⚡ The Briefing — 60 sec

• Bernie Sanders’ AI ‘gotcha’ video flops, but the memes are great No comment 😄. The clip was meant to be a clean AI takedown, but mostly ended up exposing something more useful: chatbots are still dangerously easy to steer, flatter, and bend into saying what the user already wants.

• Here are Elon Musk’s most epic goals for his vast Terafab chip plant Chips in space, because apparently Earth was not enough. Musk’s Terafab vision spans Tesla, SpaceX, and xAI, with Reuters reporting the Austin factories are intended to produce chips for vehicles, humanoid robots, and AI data centers in space.

• Tech Bytes: OpenAI dangles guaranteed returns in high-stakes enterprise AI land grab Seventeen and a half percent is not subtle. If OpenAI is really dangling guaranteed minimum returns to lock in enterprise distribution, this is not just a product war anymore — it is capital, channel, and installed-base warfare.

🛠️ The Playbook — The AI Procurement Reality Check

Mission
Pressure-test AI vendors, model claims, and commercial terms before your company mistakes hype, channel strategy, or financial engineering for real capability.

Difficulty
Intermediate

Build time
3–4 hours

ROI
Better buying decisions, less vendor fog, and fewer expensive AI bets made on branding instead of substance.

0) Why This Matters

The AI market is getting weird in a very specific way.

On the product side, the Bernie/Claude episode was a reminder that models can still be pushed into saying what the user wants, especially when the framing is leading and the chatbot gets too agreeable. That makes demos, screenshots, and staged “proof” far less trustworthy than they look.

On the infrastructure side, Musk’s Terafab pitch shows how serious players are trying to control the stack itself — chips, fabs, packaging, robotics, vehicles, orbital compute, the lot. Reuters says Tesla and SpaceX plan two advanced chip factories in Austin, one for Tesla vehicles and Optimus, the other for AI data centers in space.

And on the commercial side, OpenAI is reportedly offering private equity firms joint-venture terms with a 17.5% minimum return, downside protection, and early model access to accelerate enterprise rollout. That is not normal software sales behaviour. That is market capture behaviour.

So the operator move is simple:

• verify what the product actually does

• verify what sits underneath it

• verify what commercial incentives are shaping the pitch

• verify whether the workflow value is real

1) Architecture

2) Workflow

1. List every AI vendor or tool under consideration, including the exact workflow it claims to improve.

2. Test each tool on a real internal task instead of relying on demos, screenshots, or marketing examples.

3. Record the likely underlying model, infrastructure dependencies, and where the vendor may just be packaging someone else’s capability.

4. Review pricing, credits, guarantees, and any commercial structure that may be artificially accelerating adoption.

5. Score output quality, reliability, and human correction load on live work.

6. Classify the tool as buy, negotiate, monitor, or reject.

3) Example Prompts

Vendor Due Diligence Prompt

You are reviewing an AI vendor before procurement.

For the product below, assess:
- likely underlying model or dependency stack
- whether the product appears to be a wrapper, workflow layer, or original capability
- what claims need validation
- what commercial terms raise concern
- the top 5 procurement risks

Product:
[insert product]

Live Workflow Test Prompt

You are evaluating whether an AI tool is actually useful in production.

Review the workflow below and estimate:
- where the tool should improve speed
- where quality may break down
- where human correction is still required
- whether the workflow is good enough for a pilot

Workflow:
[insert workflow]

Commercial Reality Prompt

You are reviewing the commercial terms of an AI partnership.

Assess:
- pricing complexity
- guaranteed return structures
- credits or subsidies
- switching-cost risks
- whether the incentives suggest weak natural demand

Return:
1. summary
2. top risks
3. questions to ask before signing

Evidence Check Prompt

You are reviewing an AI demo or transcript for false confidence.

Check whether:
- the user is leading the model
- the model is becoming overly agreeable
- the output looks persuasive without being reliable
- the example would hold up in a real workflow

Return 4 bullet points only.

4) Guardrails

• Never buy from the demo alone.

• Treat unusually generous commercial terms as a signal, not a bonus.

• Separate workflow value from model brand.

• Verify whether the vendor owns the core capability or packages someone else’s.

• Count human correction time as part of product quality.

• Re-test any tool that depends on long conversations or heavy personalization.

5) Pilot Rollout — 3 hours

1. Pick three AI tools your team is considering or already paying for.

2. Define one real workflow test for each tool.

3. Run each tool against the same task set and record output quality and correction load.

4. Capture pricing, credits, incentives, and any hidden dependency stack.

5. Score each tool on usefulness, transparency, and lock-in risk.

6. Use the results to decide which tools to expand, renegotiate, or cut.

6) Metrics

• Cost per workflow completed

• Human correction time per output

• Vendor transparency score

• Number of hidden dependencies discovered

• Pilot pass rate across live workflows

• Contract complexity score

• Percentage of AI tools moved to expand, negotiate, or reject

Pro Tip: In AI procurement, the weirdest deal term in the room is usually telling you more than the slickest product demo.

🎯 The Arsenal — Tools & Platforms

• Airtable · simple vendor and workflow tracking layer for procurement reviews · Airtable

• Google Sheets · lightweight scoring model for pricing, quality, and lock-in risk · Google Sheets

• ChatGPT / Claude · useful for benchmarking vendors and pressure-testing claims, but only if you evaluate them on real work · ChatGPT · Anthropic

• TechCrunch · strong signal source for vendor, model, and platform reality checks · TechCrunch

• Reuters · useful for filtering the commercial theatre out of major AI infrastructure announcements · Reuters

Copy-paste prompt block:

You are helping me run an AI Procurement Reality Check.

For the tool or vendor below:
1. identify the claimed use case
2. identify the likely model or infrastructure dependencies
3. identify what should be tested in a live workflow
4. identify any unusual pricing, credits, guarantees, or lock-in terms
5. score the transparency of the vendor
6. list the top 5 procurement risks
7. recommend: buy, negotiate, monitor, or reject

Tool or vendor:
[insert name here]

Return the answer in markdown with sections for:
- Claimed value
- Likely dependencies
- Live workflow test
- Commercial risks
- Procurement risks
- Recommendation
- Metrics

💡 Free Office Hours

If you are trying to work out whether an AI product is the real thing, a wrapper with lipstick, or just a spicy commercial deal in disguise, I run free office hours to help map the workflow, the risk, and the buying decision.

Book here: https://calendly.com

🕹️ Game Over

The AI gold rush is loud. Procurement still has to be sober.

— Aaron Automating the boring. Amplifying the brilliant.

Subscribe: link