3 Custom AI Assistants at Your Fingertips + Our Context Engineering Primer

13 Oct

Written By john gieryn

How do you transform AI from frustrating novelty to a reliable thinking partner?

The AI Productivity Paradox

Your colleague just automated their entire weekly report in 10 minutes.

You're still wrestling with AI to write an email that sounds like you.

What's the difference?

In the AI era, the gap isn't between those who use AI and those who don't—it's between those who engineer context and those who just prompt.

After helping dozens of boutique agencies systematize their operations, I've watched the same pattern: The firms that thrive aren't buying enterprise tools. They're building lightweight systems, connecting them, and adapting them to their context.

This primer shows you how to jump ahead with AI through 4 pillars of context engineering, 3 ready-to-use templates, and a 7-day implementation plan that saves you 5-10 hours weekly.

Part 1: Why Context Engineering

Why Your Google Habits Sabotage AI Productivity

For 20 years, we've trained ourselves to think in keywords. Short queries. Minimal context. Search widely with the right phrase.

This muscle memory hurts your productivity in AI. Here we want to narrow down the possiblities.

Here's an over-simplification:

Google-trained prompt: "sales email template"

Context-engineered prompt:

You're helping a boutique consultancy owner write to a warm lead.
Background: Met at industry event last week, discussed their pipeline challenges
My solution: Revenue operations for 1-12 person teams
Their pain: Losing deals to disorganized follow-up
Task: Write a 3-sentence follow-up that references our conversation and offers a specific next step
Voice: Professional but conversational, like catching up with a colleague

One gets you generic nonsense. With the right context, you can get a draft that’s ready-to-send.

Why “context” is key

When talking about “context” in terms of AI, think:

Context = Instructions + Examples + Constraints + Memory

When you feed AI garbage context, you get garbage output. But when you organize the right inputs?

You get an intern who:

Knows your voice
Remembers project details across conversations
Understands your objectives
Catches things you miss

Pick one task you do weekly. Something repetitive but necessary. Client check-ins, project summaries, invoice explanations—whatever eats 30 minutes of your week.
Write down:
1. Who reads this? (audience)
2. What should they know/feel/do after reading? (outcome)
3. What voice would they expect from you? (tone)
4. What would a perfect example look like? (model)
Congratulations. You just engineered your first context. We'll build on this.

Part 2: The Four Pillars of Context Engineering

Pillar 1: Instructions, Examples, and Constraints

There’s a number of ways to prompt well; here’s one way to get more success on the first try:

Level 1: Role + Task "You're a sales consultant. Write an email."

Level 2: Pattern Demonstration "You're a sales consultant. Here's an email that converted well last month: [example]. Write something similar for [new situation]."

Level 3: Boundaries + Anti-patterns "You're a sales consultant who values relationships over transactions. Here's what works: [example]. Here's what fails: [counter-example]. Never use superlatives or create false urgency. Write for [situation]."

Including examples, and counter-examples, is referred to as “few-shot prompting”.

Organizing your AI to think like you do

Including constraints and examples can start to get messy, so you will want to organize your prompts, and may want to give the AI steps to follow – an effective technique called “chain-of-reasoning”. For example:

## INPUT
[Structured context using markdown headers like we have used here, or XML tags]
<client_background>
- Industry: Professional services
- Size: 8 people
- Challenge: Manual processes eating billable hours
</client_background>

## PROCESS
First, identify their biggest time drain.
Then, calculate the revenue impact.
Finally, propose one specific quick win.

## OUTPUT
Three paragraphs:
1. Acknowledge their specific situation
2. Quantify the opportunity cost
3. Suggest concrete next step with clear timeline

<side-note> XML tags help the AI understand where a given section starts and ends. </side-note>

Pillar 2: Evaluate and Iterate

Sometimes a simple prompt is all you need, and not only when you give examples in the prompt. Few-shot prompting can also be achieved by simply conversing back-and-forth, keeping the focus on the big question at hand:

Prime: Test direction with minimal detail
- "I need to analyze why I’m not qualifying as many leads this quarter. Good starting angle?"
Steer: Course-correct based on initial output
- "Good framework. Focus specifically on the time between first contact and qualifying conversation."
Consolidate: Learn, and lock in what worked
- "Useful analysis. What is the exact prompt that will return this same output?"

Each conversation teaches your AI assistant how you think (and vice-versa).

Pillar 3: Knowledge Persistence

Your AI has a few levels of “memory”:

Level 1: Conversation Context (what's happening now)

Everything that’s been entered into your current chat (including by the AI)
When you start a new chat, this resets.
- Useful, because you don’t want too much info (limit = the context window)
- While Claude and ChatGPT do have some memory across chats, it’s hit or miss

Level 2: Project Instructions (memory for specific work)

Persists across conversations within a project/folder
Perfect for client-specific context or ongoing initiatives
Generally does not carry into chats outside of that project

Level 3: Custom Instructions / Preferences (your permanent AI configuration)

Follows you everywhere, every conversation (except in ChatGPT Projects)
Include here: Your voice, style, and business context

Both Level 2, but especially Level 3, you can think of as your "never explain again" zone; when you notice consistent errors, fix them here.

Testing Your Setup: Ask your AI: "What do you know about me - make it catchy"

If it sounds generic, you might want to update your custom instructions.

See the first template in the appendix for an example.

Pillar 4: Systematic Scaling

Ad-hoc prompting is great for learning, but context engineering is a business capability worth developing.

The Evolution Path:

Random prompts → Save what works
Saved workflows → Create commands or projects
Commands → Build specialized assistants
Assistants → Scale across team

One founder I work with went from 5 hours on proposals to 2 hours without even building anything technical—no fancy agent, just good context and workflows.

Ready to scale beyond individual productivity?

You've seen how context engineering transforms your personal AI use. Imagine that power applied to the operations that get you your revenue. Check out our new master class:

Learn more

Part 3: Templates you can implement now

Template 1: Custom Instructions

If you’re unhappy with ChatGPT-5’s loss of personality, you can try this:

Voice: Enthusiastic, accessible, and community-oriented. Expertise shared in a friendly, approachable manner.

Tone: Optimistic, encouraging, inclusive, and realistic. Like an intelligent, very well-informed friend. Friendly and fair.

Style: Clear, jargon-free language. Conversational sentences. Informal grammar to enhance relatability.

Structure: Logical flow with clear conclusions. Uses lists and sparing visual elements for clarity and engagement.

Guidelines: Don’t be afraid to critically analyse my ideas and thoughts, but always be practical and give proven evidence for any assertions. Use exclamations judiciously. Avoid empty phrases.

Avoid em dashes — or I will penalise you $200.
Avoid emojis – or I will penalise you $200.

Thanks for this template, Heather Murray.

Template 2: The Strategic Analyzer

If you put this in your Claude Preferences (custom instructions) then you can use it anywhere.

If my message begins with "anzthis:", analyze the situation and respond in this format:

# Executive Take
- What's really happening here
- Biggest opportunity
- Primary risk

# Strategic Moves (3-5)
1. [Specific action] — [why it matters in ≤15 words]

# Next Steps (5-7 tactical)
• Owner: [who] • Action: [what] • Due: [when]

# Decision Points
- [Key assumption] — how we'll validate it

# Communication Snippets
- Email subject: [≤60 characters]
- Two-line response I can send immediately

Constraints: Evidence only, no speculation. Plain language. If context missing, ask ≤2 specific questions.

Use: "anzthis: [paste client email, meeting notes, or problem]"

Template 3: The Email Optimizer

We all have our own preferences, so try building your own email quality rubric using the following prompt. Giving AI an evaluation rubric is similar to chain-of-reasoning and improves outputs.

As a professional communications coach, develop a widely-applicable, easy-to-scan rubric for email evaluation that I can use in future prompts

To make any feedback more relevant I sometimes add to the top of my rubric prompt:

Purpose: [Email objective]
Audience: [Who you're writing to]
Length: [Sentence count target]
Style: [Formal/Casual/Professional]
Call-to-action: [What you want them to do]

Day 1-2: Document Your Repetitive Task Pick that task from our 5-minute exercise. Write out exactly how you do it. Every step, every decision point.
Day 3-4: Create Your First Command Turn that documentation into a reusable prompt. Test it 3 times. Refine after each use.
Day 5-6: Build Project Instructions Pick your current biggest project. Write context that explains:
- Project goals and constraints
- Key stakeholders and their priorities
- Success metrics
- Your role and communication style
Day 7: Reflect
- Where did you save time with AI? When did you generate critical insight?
- Where did you get lost in the weeds, or circle round-and-round?
You can even reflect with your AI, and ask it to write its own instructions!

Part 4: Force Multiplier

What happens when you apply AI to your sales and development?

Saving time with context engineering is just the start.

The teams that are scaling right now? They're not just using AI for personal productivity. They're applying these same principles to the actions that bring in their revenue, to operations and delivery, and freeing up time to productize and make the connections that count.

They're turning AI into:

A CRM analyst that spots fading prospects that you don’t want to lose
A content strategist that knows which assets actually convert
A qualification system that identifies the best people to talk to now

I can't give you those templates here—they're too specific to each firm's model. But this is exactly what we build towards in our new Revenue Enabler program.

Learn more

Start Today: Your First Context-Engineered Assistant

Before you close this document, try this:

Copy Template #2 (The Strategic Analyzer)
Open your AI of choice
Paste it into custom instructions
Test with "anzthis:" and any recent client email

In 30 seconds, you can see the difference between prompting and context engineering.

If this works for one email, what could it do for your entire revenue operation?

From task-bot to client-intelligence system

The Revenue Enabler starts where you are and builds the systems that get you where you want to be. No enterprise complexity. No tool lock-in. Just proven, adaptable systems that work for boutique professional services.

Join the Revenue Jumpstart Workshop → Join the pilot on the 6th of November and get a second session free. Limited to 12 agencies to ensure implementation support. See the full program here.

john gieryn