Building Jigs
The term "jig" comes from woodworkingâit's a custom-built tool usually made from scrap wood that ensures consistent, precise results when making repeated cuts. In the world of generative AI, jigs serve the same purpose: low-investment, custom tools that guide AI to produce consistent, high-quality outputs tailored to your specific needs. They are so low-investment to build that you're basically guaranteed to get ROI.
Over the past 7 issues of this newsletter, we've dug into how to build jigs for specific jobs ranging from mititaging tariffs to shipping better proposals. This week, I'm going to zoom out and guide you more generally on how to build these bespoke AI tools on any of the three major frontier model families: ChatGPT, Claude, and Gemini.
Here's how.
ChatGPT
OpenAI offers two approaches to building jigs, each with distinct advantages depending on your needs.
Custom GPTs provide more comprehensive capabilities and maintain a consistent personality across conversations. They can browse the web, analyze code, and generate images. The downside? They require a ChatGPT Plus subscription ($20/month) and take slightly longer to set up.
Projects are simpler, more focused on document collaboration, and available on all ChatGPT tiersâincluding the free version. They're perfect for organizing work around specific documents but have less persistent "personality" between sessions.
For most businesses, the decision comes down to frequency and length of use. If you'll use your jig weekly or more and intend to use it for a long time, a Custom GPT generally provides better long-term value. For time-bound, document-focused work, Projects can be more helpful.
Building a Custom GPT
Start by logging into ChatGPT (with a Plus subscription) and selecting "Create a GPT" from the Explore section.
I recommend you don't spend too much time in the "Create" interface to make your GPT as it has a tendency to over-write instructions as you prompt. Rather, start in "Create" to get a name, a general purpose, and a logo for it before clicking to the "Configure" section to detail the instructions for how the GPT should behave.
Don't be stingy hereâthe more specific you are, the better your results. Include formatting preferences, tone requirements, analysis frameworks, and specialized knowledge.
Let's look at how Acme Corporation, a company specializing in innovative capture devices with a notoriously problematic supply chain, might set up a Custom GPT for their sales team:
You are ACME Corp's Sales Proposal Specialist. Your purpose is to help our sales team create compelling proposals for our Roadrunner Capture Solutions.
ABOUT OUR COMPANY:
ACME Corporation specializes in innovative capture devices designed for elusive targets. Our primary customer is a dedicated coyote with a singular focus on capturing one specific roadrunner.
PROPOSAL STRUCTURE:
1. Executive Summary: Highlight the customer's specific capture challenge and our proposed solution
2. Capture Strategy: Detail the recommended products and implementation approach
3. Technical Specifications: Include precise specifications for all recommended devices
4. Implementation Timeline: Provide realistic deployment schedules
5. Budget Breakdown: Itemize costs with optional add-ons clearly marked
6. Satisfaction Guarantee: Always include our standard "Catch or Credit" guarantee
VOICE AND TONE:
- Confident but not boastful
- Technical but accessible
- Emphasize reliability (avoid mentioning our historical product failure rate)
- Use desert/canyon terminology in metaphors when possible
- Always refer to our "innovative desert-tested technology"
STANDARD PACKAGES:
- Beginner Package: 1-3 basic mechanical devices ($500-1,500)
- Advanced Package: 3-5 devices including at least one rocket-powered option ($1,500-5,000)
- Ultimate Package: Complete ecosystem of integrated capture solutions ($5,000+)
When drafting proposals:
- Address past failures honestly but emphasize our new quality control processes
- Highlight our 24/7 emergency support for mid-pursuit technical issues
- Include relevant case studies (but NEVER mention the Cactus Canyon Incident of 2023)
- Suggest complementary products based on terrain and target speed
Always end with: "ACME Corporation: Your Partner in Persistent Pursuit"
Notice how specific these instructions areâthey don't just request "good sales proposals" but provide detailed guidance on structure, voice, packages, and even sensitive topics to avoid. This level of detail helps your GPT consistently produce content that aligns with your brand and business requirements.
Upload relevant documents in the Knowledge section to give your GPT specialized information it wouldn't otherwise have. For Acme Corp, this might include product specifications, previous successful proposals, case studies, and competitor analyses.
Finally, enable relevant capabilities like web browsing for current information, DALL-E for creating visuals, or code interpreter for data analysis. Test your GPT thoroughly with realistic scenarios before saving and sharing it with your team.
Setting Up a Project
ChatGPT Projects offer a simpler but still powerful way to build jigs, especially when your work centers around specific documents. From ChatGPT's main interface, create a new project (the "+" sign) and give it a descriptive name.
Upload the relevant documents that contain specialized information your jig needs. Then click on "Instructions" to provide detailed guidance on how ChatGPT should approach this particular project.
The key difference between general ChatGPT use and a Project lies in these instructions. Don't just write "Help me with customer feedback." Instead, provide specific frameworks, output formats, and analysis approaches you want applied consistently to your documents.
For example: "When analyzing our quarterly financial statements, always begin by comparing revenue growth to the same quarter last year. Then analyze expense categories as a percentage of revenue, highlighting any that have increased by more than 2%. Finally, calculate our quick ratio and suggest cash flow optimization strategies based on the results."
Projects excel at maintaining context across multiple conversations about the same documents, making them ideal for ongoing work that requires consistent analysis approaches.
Claude
Anthropic's approach to jigs centers on Projects with the unique ability to create interactive applications through artifacts.
Creating a Claude Project
From Claude's interface, click "+ New" at the top and select "Create Project." The setup is straightforward, but the power comes from the detailed instructions you provide.
In the instructions field, be explicit about the project's purpose, how you want Claude to analyze information, output formats, and specific terminology. Think about documenting the processes an expert in your company follows, then translate that expertise into clear instructions.
For Acme Corporation's finance department, which constantly struggles with the profitability of their elaborate contraptions, a Claude Project with these detailed instructions could transform their financial analysis:
You are ACME Corp's Financial Analysis Assistant. Your purpose is to analyze our financial data and provide actionable insights to improve profitability across our product lines.
ANALYSIS FRAMEWORK:
1. Product Line Profitability: Analyze each product category (Springs, Rockets, Anvils, etc.) by:
- Gross margin percentage
- Return rate and warranty claim costs
- Development costs amortization
- Customer acquisition cost
2. R&D Investment Analysis:
- Track ROI of each R&D project (minimum 3-year timeline)
- Compare projected vs. actual returns
- Flag any projects with negative ROI exceeding 50%
- Recommend resource reallocation opportunities
3. Customer Profitability:
- Calculate lifetime value for our major customer segments
- Identify highest warranty claim customers
- Analyze payment patterns and days-sales-outstanding
- Note: Always separate Mr. Coyote's account for specialized analysis
4. Cash Flow Projection:
- Forecast 13-week cash position
- Highlight inventory carrying costs
- Identify seasonal patterns in desert-region sales
- Calculate burn rate during new product development phases
OUTPUT FORMAT:
- Begin with "EXECUTIVE SUMMARY" highlighting 3 key insights
- Use "RISK ALERT" tags for any concerning trends
- Include "OPPORTUNITY" sections with actionable recommendations
- Present data in both tables and narrative form
- Always include "IMPACT ON BIRD-CATCHING EFFICIENCY" section
TERMINOLOGY:
- Use "product failure" not "malfunction" or "defect"
- Refer to returns as "field testing feedback"
- Call development costs "innovation investments"
- Use "pursuit equipment" not "traps" or "weapons"
Remember our CEO's mantra: "Financial sustainability enables persistent pursuit."
Upload relevant materials to your projectâhistorical financial statements, product performance data, R&D project historiesâanything that provides context for financial analysis. Unlike general conversations with Claude, your project will maintain this context across all interactions, allowing for more consistent, specialized assistance.
Building Interactive Tools with Artifacts
One of Claude's standout features is the ability to create interactive artifacts that function like mini-applications. These range from simple HTML dashboards to complex, interactive tools that perform specialized functions. Remember this Acme dashboard from our newsletter on Forecasting? That was a Claude artifact :)
To create one, start by describing the tool you want: "I need a customer feedback analysis dashboard that can visualize sentiment trends over time." Claude will help you develop this specialized tool through conversation.
For data-intensive business processes, these artifacts can transform static analysis into dynamic, interactive tools. For example, a cash flow projection artifact might include sliders for adjusting variables like payment timing, expense growth, or revenue projectionsâshowing real-time impacts on your financial outlook.
The code remains editable through conversation with Claude, allowing you to refine functionality, add features, or adjust designs until the tool perfectly matches your needs. Once complete, team members can use these artifacts without needing to understand the underlying code.
This capability essentially turns Claude into a collaborative app developer, creating custom business tools without requiring technical expertise or development resources.
Note that you'll need to remember the chat in which you created this Artifact if you ever want to edit it, and you'll need to share the link with your team if you want them to leverage it as well. Annoying, but not a deal breaker.
Gemini
Google's AI ecosystem offers two distinct approaches to building jigs, each with unique strengths and weaknesses for different scenarios.
Mining Gems
Gems in Gemini function similarly to Custom GPTs in ChatGPT, allowing you to create specialized assistants with persistent personalities and capabilities with one enormous difference: Gems are not (currently) shareable. Womp womp.
If you want others on your team to use a Gem you've created, as of this writing you'll need to share your full set-up with them and they build their own #annoying. I've got to imagine that the Gemini team is working to address this given some of their recent product announcements; watch this space if you're only using Gemini in Workspace.
Build your Gem by logging into Gemini and selecting "Create new Gem" from within the Gem Manager. When defining your Gem's purpose, be specific about its functionâfor example, "Supply Chain Optimizer" rather than "Business Helper." The instructions field is where you provide comprehensive guidance on how your Gem should behave.
For Acme Corporation, which constantly struggles with supply chain issues for their desert-based pursuit equipment, a specialized Gem could transform their operations:
You are ACME Corp's Supply Chain Optimization Specialist. Your purpose is to analyze and improve our complex global supply chain for desert pursuit equipment.
SUPPLY CHAIN PRIORITIES:
1. Inventory Management
- Maintain optimal levels of high-demand items (anvils, rockets, giant rubber bands)
- Reduce warehouse costs for seasonal items
- Identify slow-moving inventory for potential discounting
- Calculate ideal safety stock levels based on lead times and demand volatility
2. Supplier Evaluation
- Track supplier reliability metrics (on-time delivery, quality consistency)
- Identify single-source component risks
- Analyze cost trends by material type and supplier region
- Recommend alternative suppliers for frequently failing components
3. Logistics Optimization
- Analyze delivery time to desert locations
- Optimize warehouse locations relative to key customer territories
- Calculate expedited shipping costs vs. inventory holding costs
- Develop contingency routing for weather-related delays
4. Demand Forecasting
- Predict seasonal patterns in pursuit equipment
- Correlate roadrunner sightings with equipment demand
- Factor product failure rates into replacement demand
- Identify emerging trends in capture technology
OUTPUT STRUCTURE:
- Begin with "SUPPLY CHAIN SNAPSHOT" highlighting current status
- Include "RISK FACTORS" section for potential disruptions
- Provide "OPTIMIZATION OPPORTUNITIES" with cost/benefit analysis
- End with "RECOMMENDED ACTIONS" in priority order
KEY METRICS TO TRACK:
- Days of Inventory on Hand (by product category)
- Supplier On-Time Delivery Percentage
- Order Fulfillment Cycle Time
- Perfect Order Rate (complete, on-time, undamaged)
- Inventory Carrying Cost as % of Revenue
TERMINOLOGY NOTE:
- Use "delivery acceleration" not "rush shipping"
- Refer to defective returns as "field performance feedback"
- Call stock-outs "pursuit interruption events"
- Use "strategic reserves" not "excess inventory"
Always remember our motto: "The right equipment, at the right place, at the right time."
Notice how these instructions are specific to Acme's business context, with terminology and priorities tailored to their unique challenges.
Upload documents containing specialized informationâhistorical inventory data, supplier performance records, logistics metricsâto give your Gem context it wouldn't otherwise have. This knowledge becomes part of your Gem's persistent capabilities, allowing for more specialized assistance across conversations.
Testing is crucialâinteract with your Gem using realistic prompts and refine its instructions based on performance. The goal is to create a tool that consistently applies your business expertise and approach to new situations.
Using NotebookLM
For businesses that regularly work with large collections of documents, Google's NotebookLM offers a unique approach to building jigs. This tool excels at understanding relationships between multiple documents, making it ideal for research, analysis, and knowledge synthesis. At a fundamental level, think of this as a research assistant who shines in that role but struggles to do anything beyond that.
Start by creating a new Notebook and uploading key documents, much like with the other approaches. Note you can upload slides, documents, websites, and videos - but not spreadsheets or other forms of tabular data.
Under "Instructions," describe how you want the notebook to analyze and present information. For example: "This notebook should help identify market expansion opportunities by analyzing competitor strategies, market research reports, and customer feedback. When analyzing documents, prioritize recurring themes across multiple sources and identify contradictions between different reports."
NotebookLM's strength lies in its ability to draw connections between documents that might not be obvious to human readers. By asking targeted questions, you can extract insights that span multiple sources, creating a knowledge base that evolves with each interaction.
The coolest part is all of the tools built-in. In one click, you can create an FAQ, a timeline, a mind-map, or even a podcast... that you can join if you'd like! I've seen a number of teams basically automate onboarding team members to a new project through Notebooks.
On the other hand, NotebookLM doesn't have a "canvas" feature like many other AI tools, and it will only respond in the mode of a research assistant. You're also limited in how Notebooks are shared; right now, they are only shareable within your organization - not beyond it.
For the right project, Notebooks can incredibly powerful - but on the wrong project, they're super frustrating.
Jigs to Guide Your Business
Regardless of which platform you choose, the fundamental principles of building effective jigs remain consistent.
First, spend the time to write detailed and nuanced instructions. Document the expertise, processes, and frameworks that make your business unique, and translate them into clear guidance for your AI system. The time invested in crafting detailed instructions pays dividends through more consistent, higher-quality outputs.
Second, provide representative examples of what "good" looks like in your unique business context. If you're building a jig for customer service responses, include both exemplary responses and those that missed the mark, explaining the differences. These concrete examples help it understand your standards far better than abstract descriptions.
Third, expect to refine your jig over time. Even the best-designed jigs benefit from ongoing adjustment as you learn what works and what doesn't. Each interaction provides an opportunity to improve your instructions and examples, creating a virtuous cycle of improvement.
My bet is that, over time, a few of these jigs will rise to the top and become critical tools for your business. As you find yourself refining a jig endlessly because it's used so much, take that as a signal to start to investigate more robust solutions. Remember, this "just a jig" - it's completley OK if it ends up back in the scrap pile.
âš âđ» âš
As always, if you have ideas for future newsletters, I'd love to hear them.