Composable Architecture: A Phased Roadmap to Get It Right
Composable architecture roadmap from customer 1 to 20+
TABLE OF CONTENTS
Constraint-Based Prompting for Real-World AI Projects
A 5-step guide to constraint-based prompting for real accuracy
This is part of a series about Innovation Strategy
List view
Why Do AI Projects Fail? 3 Fixes That Actually Work
Composable Architecture: A Phased Roadmap to Get It Right
Decision Memo Template: A 5-Step Roadmap to Fast Executive Approvals
Break the Cycle of Automation Fatigue for Good
Overengineering: Stop Feeding Wants and Starving Your Needs
Asymmetric Strategy: Shogun's 3 Steps to Win Uneven Fights
Getting Your First Customers in Five Hours a Week
How to Break Your Growth Ceiling While Bootstrapping a SaaS
Fast Follower Strategy: 4 Startup Idea Shortcuts
What The Lord of the Rings Teaches Us: 3 Powerful Strategy Lessons
Problem Validation: Solve a Real Problem Before It’s Too Late
Feeling Overwhelmed? Here's How to Prioritize Strategic Initiatives
Assumption Testing: A Strategic Framework for Success
What is a Calculated Risk? An Essential Guide
The Regret Minimization Framework: A Practical Checklist
Strategic Prioritization: Why Your First Move Might Be Your Biggest Mistake
Strategic Decision Making Framework: Don’t Miss This Essential Tool
Problem Statement Worksheet: 30 Minute Exercise to Clarity
This Problem Statement Template Delivers Results
How to Write a Problem Statement: Avoid These Common Mistakes
Tech-Enabled Services Versus SaaS Business Model: Key Metrics
Constraint-Based Prompting for Real-World Projects
Most AI advice fails the moment it meets your real project. You ask for a fix, follow it exactly, and watch it fall apart against your real budget, staff, or land. If you keep asking AI questions without a system, you will keep losing time and money on advice built for someone else's setup.
But if you surface with AI your real limits first, every answer already fits your situation. This is constraint-based prompting. It stops guessing. Then starts building something that can survive.
Why this matters
Here's the part that catches most people off guard. You ask AI for a fix, and it hands you a confident, specific answer, its top pick, ranked above everything else. Then you either ask a follow-up question, or you actually try the plan, and a detail comes up that never got mentioned before: a condition the AI never asked about and you never thought to state.
Once that one detail enters the conversation, the AI's own top pick can drop straight to the bottom of the list, and a completely different option jumps to the top.
Nothing about your project changed. The only thing that changed is that a hidden requirement finally got said out loud.
The 3 Core Drivers Behind Weak Constraint-Based Prompting
Most people fall into one of two habits when they ask AI for help.
The first is the raw question: asking something broad, like "What's a good volunteer program model?" and hoping the AI flags any hidden risks on its own.
The second is the general-context question: adding your city or your field, but leaving out the real limits underneath.
Both habits lead to the same three problems, and each one maps to one of the three steps below.
Before we dive in, our running example
Marcus runs a small nonprofit in Sacramento that trains people who've struggled to find steady work for warehouse and logistics jobs. He also owns an acre of land outside the city that he wants to turn into a low-care, fire-safe yard. Both projects taught him the same hard lesson about AI advice, so we'll follow his nonprofit as the main story and check in on his yard along the way.
Default Optimization Bias
When you don't state your limits, AI picks the easiest, most common answer. Marcus once asked for a way to boost volunteer turnout, and the AI suggested a weekly phone check-in run by a coordinator. It sounded great, but Marcus has no paid staff for that role. His yard project had the same problem: the AI once called a groundcover "60% more water efficient" than a normal lawn, without asking whether Marcus had any water at all to give it. Step 1 fixes this.
Missing Foundation Data
Generic advice focuses on what looks good or works fast. It doesn't ask whether a plan can survive your actual starting point, or whether you're even scoring it against the right criteria. Marcus's first outreach plan used only one channel, social media, and left gaps a competing program quickly filled. His yard had the same problem: one type of groundcover left bare patches of dirt that invasive weeds moved into. Step 2 fixes this.
Delayed Failure Feedback
A plan can look great for weeks and still be doomed once real stress hits, especially if you only considered one option instead of a wide set. Marcus's first training group had strong turnout for three months, then attendance dropped fast once the early excitement faded. His yard told the same story: grass planted in March sprouted fast, then rotted by June because its roots never got a full season to grow. Step 3 fixes this.
The 3-Step Constraint-Based Prompting Method
Each step below closes exactly one of the three gaps above. Each one also comes with a reusable prompt you can copy, paste, and adapt to any project, whether that project is a nonprofit program, a business plan, or a backyard.
Step 1: Name the Real Problem and Your Real Limits
What This Is
Before asking AI for a single option, ask it to state back what it thinks your real underlying problem is. Then tell it your hard limits: what money, staff, time, or conditions you do not have. This does two things at once. It checks whether the AI actually understood your situation, and it locks out any answer built for a better-resourced version of your project.
Why It Matters
This closes the default optimization bias gap. Without a stated problem and stated limits, AI defaults to the easiest common setup. It hands you a plan built for ideal support, not your reality.
The Reusable Prompt
"Before you suggest anything, tell me what you think the real underlying problem is here, based on what I've told you. Then list back the hard limits I'm working with. Rank every future option against success under these limits first, not just ease."
Examples (Toggle for more)
- Less Productive
- Marcus asks, "What's a good way to run a volunteer mentor program?" The AI suggests a weekly check-in system run by a full-time coordinator, a role Marcus never mentioned he didn't have.
- More Productive
- Marcus runs the Step 1 prompt and the AI states back the real problem: keeping volunteers engaged without any paid staff to manage them.
- Marcus confirms that's right, then adds his real limits: zero paid coordinator time and zero guaranteed funding past this quarter.
- Decision and result: the AI drops every plan that needs ongoing staff time and suggests a peer-led check-in system where trained volunteers support each other in pairs.
- At home: Marcus runs the same prompt for his yard. The AI states the real problem back as "a fire-safe ground cover that survives with no watering," and Marcus confirms that's exactly right before any options come up.
Step 2: Surface Hidden Risks and Build a Complete Scorecard
What This Is
Ask the AI to name the risks and hidden assumptions that could sink your plan later. This includes bad timing. Then do a second check. Hand the AI your list of decision criteria. Ask it to check whether that list is complete and correctly weighted, before it looks at a single option.
Why It Matters
This closes the missing foundation data gap. A plan can fail either because a real-world risk was never named, or because you were scoring options against the wrong criteria from the start. Fixing the scorecard before you shop for options prevents disappointment down the line.
The Reusable Prompts
"Before recommending anything, name the top five hidden risks or assumptions that would cause this plan to fail if left unmanaged. Include timing risk: what's the hardest stretch ahead, and would this plan survive it?"
"Here are my top criteria: [list them]. Do not evaluate any options yet. Instead, score 0-100% whether this is the correct and complete set of criteria, and tell me if I'm missing any critical criteria that would lead to disappointment or failure later. Then tell me if any criteria should be weighted higher, like 1.5x or 2x, and why."
Examples (Toggle for more)
- Less Productive
- Marcus accepts a plan to recruit trainees through a single employer partnership without asking what could go wrong. Six months in, that employer changes managers and drops the partnership, and the pipeline dries up overnight.
- More Productive
- Marcus runs the risk prompt, and the AI flags reliance on one employer contact as a single point of failure, plus a timing risk: launching right before the holidays, when attendance always drops.
- He then runs the criteria-completeness prompt with his list: cost, trainee turnout, employer interest. The AI scores this 60% complete and flags a missing criterion, backup hiring paths if one employer slows down, and suggests weighting "backup hiring paths" at 2x since a single point of failure caused his last program's biggest loss.
- Decision and result: Marcus adds the missing criterion before he even starts comparing options, so nothing critical gets scored as an afterthought.
- At home: Marcus runs the same two prompts on his yard plan. The AI flags seed-eating birds as a hidden risk he hadn't considered, and flags that his criteria list is missing "survives without a full growing season," which becomes his highest-weighted criterion.
Step 3: Brainstorm Wide, Then Score Against the Scorecard
What This Is
Once your problem, limits, risks, and criteria are locked in, ask the AI for a wide set of options. They should be genuinely different from each other, not small variations on the same idea. Then have the AI score every one against your finished, weighted scorecard.
Why It Matters
This closes the delayed failure feedback gap. Betting on the first plan that sounds good, or only comparing two or three similar options, is how a plan looks fine for weeks and then fails once real stress hits. A wide set of genuinely different options, scored honestly, surfaces the ones built to last before you commit.
The Reusable Prompt
"Using the criteria and weights we just finalized, brainstorm 30 diverse, non-overlapping solution categories. Don't give me small variations on one idea, provide distinct categories and examples within each. Score each one against the weighted criteria. Flag which options cover each other's weaknesses if combined."
Examples (Toggle for more)
- Less Productive
- Marcus compares only two outreach ideas: a flyer campaign and a single employer partnership. Both share the same weakness. They depend on one channel with no backup.
- More Productive
- Marcus runs the Step 3 prompt with his finished, weighted criteria. The AI returns 30 genuinely different outreach and hiring-pipeline ideas, from community radio spots to a peer-referral bonus system to a shared hiring pool with two other nonprofits.
- Each idea gets scored against his weighted criteria, and the AI flags that a single employer partnership and a staffing-agency partnership cover each other's biggest weaknesses if combined.
- Decision and result: Marcus builds a plan with three combined, non-overlapping paths to a job for his trainees, instead of one fragile bet.
- At home: Marcus runs the same prompt for his yard. The AI returns 30 different groundcover and planting-timing combinations instead of one plant, and flags that pairing a clump-forming grass with a spreading variety and a low seed-eating-risk plant covers three weaknesses at once.
Before this approach, an open-ended AI search led Marcus toward plans that looked strong on paper. They failed within months. One had a training pipeline with a single point of failure. Another was a yard planted at the wrong time with no backup plants. Both routes cost him real time, money, and trust. After using this three-step approach, things changed. His training program keeps running through slow seasons because it has more than one path to a job. His acre survives the driest, hottest months without watering, because the plan was scored against the right criteria from the start, not a best-case guess.
Constraint-Based Prompting Actionable Checklist and Toolkit
Checklist (Toggle for more)
- "Before you suggest anything, tell me what you think the real underlying problem is here. Then list back the hard limits I'm working with. Rank every option against survival first, not ease." (use for Step 1)
- "Before recommending anything, name the top three hidden risks or assumptions that would cause this plan to fail if left unmanaged, including timing risk." (use for Step 2)
- "Here are my top criteria: [list]. Score 0-100% whether this is complete, flag anything missing given my context, and tell me if any criteria should be weighted higher, like 1.5x or 2x." (use for Step 2)
- "Using the criteria and weights we just finalized, brainstorm 30 diverse, non-overlapping solution categories and score each one against the weighted criteria." (use for Step 3)
Toolkit (Toggle for more)
- Constraint-Based Prompting Action Plan — the master checklist that walks you through all three steps in order, so you don't skip one.
- Problem and Limits Prompt — confirms the AI understands your real problem, then locks in your hard limits before any options appear.
- Risk and Criteria Scorecard — surfaces hidden risks and timing threats, then checks your decision criteria for completeness and correct weighting.
- Wide Brainstorm and Score — generates a genuinely diverse set of options and scores all of them against your finished scorecard at once.
Constraint-Based Prompting FAQ
What is constraint-based prompting?
It's a repeatable set of prompts you give AI before asking for advice. It stops the AI from assuming ideal conditions. Instead, it works from your real problem, your real limits, and a complete, correctly weighted set of criteria.
Do I need to be technical to use this?
No. Every step here is a single copy-paste prompt, not code or special software.
How is this different from a normal prompt template?
Most templates focus on tone and clarity. This method focuses on three things most people skip: confirming the AI understands the real problem, checking your criteria for gaps before you shop for options, and comparing a wide set of choices instead of just one or two.
Can I use this for things besides a nonprofit program or a yard?
Yes. The three steps apply to any real-world plan, from a small business rollout to a home project, anywhere AI might assume support you don't actually have or skip straight to options before checking your criteria.
When should I not use this?
Skip it for quick, low-stakes questions, like picking a paint color, where a wrong answer costs you nothing. Use the full method when a bad plan would cost real money, time, or trust.
Dan Wu, JD/PhD
Lead Innovation Advisor
I build and advise mission-driven ventures to scale like startups.
SVP of Product & Chief Strategy Officer.
- As a go-to-market-focused product leader, I’ve led and launched products and teams at tech startups in highly-regulated domains, ranging from 6 to 8 figures in revenue.
- Led core products and product marketing key to pre-seed to D raises across highly-regulated industries such as data/AI governance, real estate, & fintech; rebuilt buyer journeys to triple conversion rates; Won Toyota’s national startup competition.
Harvard JD/PhD focused on responsible innovation for basic needs.
- Focus on cross-sector social capital formation, with a strong background in mixed-methods research.
- Selected as a National Science Foundation fellow & published on responsible strategy and innovation in outlets like Oxford University Press, Fast Company, and TechCrunch.
First-generation college student prioritizing inclusion and belonging in his practice.
- I was raised by a single mother without a high school degree.
- I’m passionate about mentoring and coaching using methods that “works with” (versus “do to”), sensitive to one’s constraints and experiences.
Made with Bullet