Three

Podcast 2 — In conversation

Speaker A: The rule of three in rhetoric and writing is one of the most robust and well-documented principles in communication. Constructions of three have a natural rhythm and completeness that pairs and longer lists do not — veni, vidi, vici; blood, sweat, and tears; life, liberty, and the pursuit of happiness. Rhetoricians in the classical tradition were aware of this — the tricolon, a sentence of three parallel elements, was considered one of the most effective rhetorical structures. Modern research on persuasion and memory retention has confirmed that lists of three are better remembered and more persuasive than lists of two or four. The human brain processes three items as a complete, balanced unit.

Speaker B: In software development, three appears as a meaningful number in several important contexts. A three-tier architecture — presentation layer, business logic layer, and data layer — is the standard model for many web applications. Three is often the minimum number of instances required for a distributed system to achieve consensus — a three-node cluster can tolerate one node failure and still have a majority. The three-way merge in version control — common ancestor, branch A, and branch B — is the standard algorithm for reconciling divergent code histories. The number three appears at the foundations of distributed systems design in ways that reflect its mathematical properties of minimum majority.

Speaker A: Three versus trinity — what is the distinction?

Speaker B: Three is the cardinal number — a count, a quantity, an abstraction. Trinity is a word that specifically means a group or set of three — particularly a sacred or specially unified group of three. In Christian theology, the Trinity refers to the specific doctrine of the Father, Son, and Holy Spirit as three persons of one God. More broadly, trinity can describe any group of three strongly associated things — a business trinity of quality, cost, and time. Three is neutral and mathematical; trinity carries a sense of unity-within-three, of the three forming a special coherent whole rather than just a count.

Speaker A: What is a common mistake about the number three in design?

Speaker B: Applying the rule of thirds mechanically without understanding its purpose. The rule of thirds in photography and design suggests placing key elements along the lines that divide the frame into thirds horizontally and vertically, rather than centring everything. The rule exists because off-centre compositions create more visual tension and interest than centred ones. The mistake is following the rule so rigidly that naturally centred compositions are forced into thirds when they would be more effective centred. Any compositional rule is a heuristic — a starting point for understanding what tends to work — not a law that overrides visual judgment.

Speaker A: Any close synonyms?

Speaker B: Trio is the musical and performance term for a group of three. Trio can also mean any group of three things or people — a trio of challenges. Triad refers to a group of three closely related things — the triad of qualities, a musical triad of three notes. Trinity as discussed. Triplet refers to three of a kind born simultaneously, three musical notes of equal length, or any group of three identical or closely matched items. Treble in British English means three times as much — a treble dose. Triple means three-fold or consisting of three parts. Three itself is the most neutral and universal; all its synonyms carry more specific connotations about the type or relationship of the three.

Podcast 3 — Prompt engineering

Speaker B: Welcome to the workshop. Today we prompt AI using the word three — building the three-tier application architectures, three-state machine managers, three-node distributed systems, rule-of-three content structures, and tripartite data validation systems that use the power of three as a structuring principle for robust and balanced technical design.

Speaker A: So three tells the AI to build the tripartite and three-tier systems — the three-layer application architectures that separate concerns cleanly, the three-state workflow managers that handle pending-active-complete state machines, and the three-point scoring systems that evaluate quality across three dimensions?

Speaker B: Exactly. It signals three-tier architecture, tripartite design, and three-component system structure. Let me start with a three-tier application setup.

Build a classic three-tier web application structure: implement a presentation tier as a REST API serving JSON to clients, a business logic tier handling all application rules and data processing, and a data tier managing database reads and writes. Enforce strict separation — the presentation tier must not directly query the database, and the data tier must not contain business logic. Define clear interfaces between tiers. Implement a health check endpoint that reports the status of each tier independently. Log requests at the boundary of each tier with the tier name, the operation, and the duration.

Speaker B: Three-tier application enforces presentation-business-data separation with strict interface contracts, provides per-tier health check endpoints, and logs requests at each tier boundary with operation and duration — the AI builds the three-tier structure, the interface definitions, the health checks, and the boundary loggers.

Speaker A: The per-tier health check that reports each tier's status independently — not just an overall application health but presentation-tier healthy, business-tier healthy, data-tier degraded — is the diagnostic granularity that allows operators to identify exactly where a problem lies: a failed database connection shows up as data-tier unhealthy, while the presentation and business tiers remain healthy, immediately narrowing the scope of investigation.

Speaker B: Independent per-tier health status as the diagnostic granularity and investigation scope narrowing tool. Now the three-state schema.

Design a three-state workflow management schema: a workflow_definitions table recording each workflow type with an ID, name, description, and the three states it uses — a waiting state name, an active state name, and a completed state name. A workflow_instances table recording each running instance with a definition ID, current state, the entity it is applied to, the state entered timestamp, the state changed by user ID, and any state-specific metadata as JSON. A state_transitions table recording every state change with the instance ID, from state, to state, changed by user ID, timestamp, and transition reason. A workflow_summary_view showing per-definition the count of instances in each state and the average time spent in each state.

Speaker B: Three-state workflow schema stores configurable state names per workflow type, instances with current state and metadata, full state transition history with reasons, and a summary view of per-definition instance counts and average dwell time per state — the AI builds the four-table schema.

Speaker A: The average time spent in each state — computed in the summary view — reveals workflow bottlenecks: if instances average two days in the waiting state and two hours in the active state before completing, the bottleneck is in the transition from waiting to active, which is where investigation and optimisation should focus.

Speaker B: Per-state average dwell time as the workflow bottleneck identification and optimisation focus tool. Next, the three-column interface.

Design a three-panel kanban-style workflow interface: display workflow items in three columns representing the three states — To Do, In Progress, and Done. Allow dragging cards between columns to change state. Show each card with the item title, assignee avatar, priority indicator, and time in current state. Allow filtering the board by assignee, priority, and date range. Show column totals at the top of each column. Highlight cards that have been in the same state longer than a configurable threshold in amber. Show a board summary bar at the top with total items, completion rate, and the count of amber-highlighted stale items.

Speaker B: Three-column kanban shows state-grouped cards with drag-and-drop state change, title-assignee-priority-time-in-state card display, filter controls, column totals, amber stale item highlighting, and a summary bar with total items, completion rate, and stale count — the AI builds the three-column board, the drag handler, the card renderer, the filter system, the stale highlighter, and the summary bar.

Speaker A: The amber highlighting of cards that have been in the same state longer than a threshold — making stale work visible at a glance — is the workflow health indicator that reveals blockages without requiring a report: a board with twelve amber cards in the In Progress column is immediately legible as a workflow with a serious blockage, and the highlighted cards identify exactly which items are stuck.

Speaker B: Amber stale highlighting as the at-a-glance workflow blockage visibility mechanism. On to three-point scoring.

Create a three-dimensional quality scoring system for content or product evaluation: score each item across three dimensions — quality, relevance, and completeness — on a scale of one to ten. Compute an overall score as a weighted average of the three dimensions with configurable weights. Show a radar chart visualising the three scores for each item. Allow comparing two items side by side on the radar chart. Flag items where any single dimension score is below a configurable minimum threshold, even if the overall score is acceptable. Export scoring data as CSV for analysis.

Speaker B: Three-dimensional quality scorer rates quality-relevance-completeness with weighted average overall score, renders per-item radar charts, supports side-by-side comparison, flags any-dimension minimums regardless of overall score, and exports CSV — the AI builds the three-dimension scorer, the weight calculator, the radar chart renderer, the comparison view, the dimension minimum flag, and the CSV exporter.

Speaker A: The any-dimension minimum flag — alerting when a single dimension score falls below a threshold even if the overall weighted average is acceptable — prevents the averaging-away of critical weaknesses: an item scoring ten on quality and ten on relevance but one on completeness may have a perfectly acceptable average; the minimum flag ensures that the critical incompleteness is visible despite the good averages in other dimensions.

Speaker B: Single-dimension minimum flag as the critical-weakness-averaging-away prevention mechanism. Now something whole.

Build a three-tier data classification and access control system: classify all data assets into three tiers — public, internal, and confidential — based on sensitivity. Enforce access controls based on classification: public data is accessible to all authenticated users, internal data requires an active employee relationship, and confidential data requires explicit role-based permission. Apply classification automatically where possible using pattern matching and data type rules. Generate a quarterly data classification inventory report showing how many assets exist in each tier and the distribution across data types and systems.

Speaker B: Three-tier data classification system classifies assets as public-internal-confidential, enforces tier-appropriate access controls, applies automatic classification by pattern and type, and generates quarterly inventory reports with tier distribution by data type and system — the AI builds the full classification and access control platform.

Speaker A: Automatic classification using pattern matching and data type rules — classifying data based on what it contains and what type it is rather than requiring manual tagging of every asset — is the scale solution that makes classification practical across large organisations: a team of people cannot manually classify millions of database rows and files; a classification engine that recognises national insurance numbers, email addresses, and financial amounts and tags them confidential automatically can.

Speaker B: Automatic pattern-based classification as the manual-tagging impossibility-at-scale solution. Last one — three-environment deployment pipeline.

Write a three-environment deployment pipeline manager: automate the promotion of code from development to staging to production environments. Require automated test passing before any promotion. Require a human approval gate before the production promotion. Track each deployment across all three environments showing the version deployed, the deployment time, the test results, and the approver. Support instant rollback in any environment to the previously deployed version. Generate a deployment frequency and success rate report per environment.

Speaker B: Three-environment pipeline automates dev-to-staging-to-production promotion with automated test gates, a human approval gate for production, per-environment deployment tracking with version-time-tests-approver, instant rollback to previous version, and deployment frequency and success rate reports — the AI builds the promotion engine, the test gate, the approval gate, the deployment log, the rollback mechanism, and the frequency reporter.

Speaker A: The human approval gate before the production promotion — requiring a person to explicitly authorise the production deployment even after all automated tests pass — is the last line of defence against automated systems deploying at the wrong time: automated tests cannot know about a planned maintenance window, a pending customer communication, or a business reason to delay; the human gate ensures a person with context makes the final decision.

Speaker B: That is the goal — one fundamental number, better-structured three-tier and three-state systems. Keep it in your kit.