1 · Evaluate a tool
15 minTake the last tool or repository someone recommended to you and run the prompt in the AI that knows your project.
Done when: you have the 4 answers and a verdict, IN or LEFT OUT, noted in the log.
Lesson 4 · 13 min
Everything with tongs: each agent's model and the tool that comes in (or not)
Every week a new repository shows up promising everything. Handle it with tongs: before installing, ask your AI whether it fits your stack and what you gain. And don't use the same model for everything: each agent can have its own, among the best, the economical and the free.
Start the lesson
In this lesson you will
- 1Three model tiers: the best, the economical and the free
- 2Four questions before you install any tool
- 3Two opinions on a sensitive topic
- 4Why the plan takes longer than the execution
Goal: choose each agent's model by cost and task, and test any new tool against your stack before installing it.
01New tool: in or out?
Tap each step.
Click each block in the diagram to see what it does.
02The model for each agent
Three tiers. The names change every month; the criteria stay.
1. The best
Coding, finance and decisions involving money call for the strongest model your plan offers.
The cost of an error here is higher than the price difference.
2. Budget
Customer service, triage and summaries run many times a day. A cheaper, faster model can handle them.
Test with 20 real cases before switching.
3. Free
There are open, free models with huge context windows, good for prototyping and for tasks with no sensitive data.
Check the provider's data policy: free sometimes means your data trains the model.
4. Two opinions
For a financial plan, ask two agents running different models for the same analysis, with the same context, and compare.
Where they agree, you gain confidence; where they disagree, that's the question you need to ask.
03Templates to copy
Run the first one in the AI that already knows your project.
Prompt · does the tool fit?
Someone recommended this tool to me: <name and link>. You know my stack (languages, services, what's already running). Answer in up to 10 lines: 1. DOES IT FIT? What language and infrastructure does it run on, and does that work with my stack? 2. DO I ALREADY HAVE IT? What I already use that does the same thing or close to it. 3. WHAT DO I GAIN? One concrete task that would get better, with a number if possible. 4. RISK? Data access, maintenance, license, dependence on third parties. Verdict: IN (<deadline> test on one task) or OUT (reason in 1 line).
Tool log
DATE | TOOL | VERDICT | REASON <dd/mm> | <name> | out | different language, repeats what the CRM already does <dd/mm> | <name> | in | 2-week test on the purchasing agent
04Practice
Use a real recommendation you received.
2 · Each agent's tier
10 minGo back to the orchestra map (lesson 3) and mark each agent's tier: best, economical or free.
Done when: every agent on the map has a tier and the reason in a few words.
05Lesson 4 on video
The lesson's mechanism explained in about two and a half minutes.
Lesson 4 on video · 2 min 30 sVideo page with script →
06Quick quiz
4 questions.
Question 1 of 4
A colleague recommends an excellent repository, but it's all in another language and repeats what your system already does. What do you do?
Correct
A good tool that doesn't fit your stack and repeats what you already have brings upkeep with no gain. The log saves you from redoing the analysis when someone recommends it again.Not that one
A good tool that doesn't fit your stack and repeats what you already have brings upkeep with no gain. The log saves you from redoing the analysis when someone recommends it again.Review: New tool: in or out?Question 2 of 4
Which agent deserves the strongest model?
Correct
Where a mistake is costly (money, code in use, a decision), pay for the best. Summaries and frequently asked questions run a lot and are fine with an economical model.Not that one
Where a mistake is costly (money, code in use, a decision), pay for the best. Summaries and frequently asked questions run a lot and are fine with an economical model.Review: The model for each agentQuestion 3 of 4
Two agents with the same context disagreed about the cash plan. Is that bad?
Correct
The second opinion is there to reveal weak points. Where they agree, you can trust it more; where they disagree, you need more analysis, yours and not theirs.Not that one
The second opinion is there to reveal weak points. Where they agree, you can trust it more; where they disagree, you need more analysis, yours and not theirs.Review: The model for each agentQuestion 4 of 4
A free model has a huge context. What should you watch out for before using it?
Correct
Free can mean the provider uses your data. Great for a prototype and for tasks with no sensitive data; not for a client's accounting.Not that one
Free can mean the provider uses your data. Great for a prototype and for tasks with no sensitive data; not for a client's accounting.Review: The model for each agentQuiz complete
You got {a} of {n} right. Get them all right and the lesson is marked as complete.Pick an option. The explanation shows up after you try.
07Summary
With tongsFour questions before installing: fit, already have, gain, risk.
Three tiersBest where mistakes cost a lot, economical for volume, free for prototypes.
Two opinionsOn a sensitive topic, compare two models with the same context.
LogWhat was left out is noted with the reason.