Clearward.
Structured Excel and Google Sheets products that help turn everyday complexity into a calmer, more manageable system.
Product requirements, AI direction & release validation
Everyday organization, built with software discipline.
HUMAN DIRECTION → CONSIDERED SYSTEMSTreat a customer workbook as software: structured data, independent checks, and a controlled release.
A workbook has to survive everyday use
Clearward creates organization systems for everyday life and small trades businesses. The technical problem is deeper than arranging an attractive spreadsheet: inputs, calculations, and presentation become difficult to maintain when they share the same cells, record identifiers change, or the only release test is opening the file once.
The products are designed for Excel and Google Sheets without macros, Power Query, or external data connections. Those constraints favor portable workbook logic and predictable customer behavior. Internal engineering material also needs to stay separate from the files a customer receives.
Direct the agents and own the release criteria
Casey defined product requirements and directed Codex and ChatGPT through workbook generation, testing, packaging, and delivery. Persistent agent instructions described the permitted work, product expectations, QA, release procedures, account boundaries, and where human takeover was required.
The assistants performed engineering work within that direction; Casey supplied the scope and judged release readiness. The project demonstrates AI-assisted product engineering and validation, rather than a claim that every workbook, formula, or supporting script was manually authored without assistance.
Separate inputs, records, calculations, and views
Structured tables use stable record identifiers. Controlled input cells are separated from formula-driven state, with data validation and configuration tables supporting the workbook’s rules. Dashboards present the result without making the presentation layer the only place that business logic lives.
The Pet Care System is a concrete example: a 13-sheet workbook covering profiles, care, expenses, settings, a dashboard, guidance, and a printable handoff. Structured tables, stable IDs, validation rules, configurable thresholds, and compatible formulas make those views part of one coherent data model rather than disconnected templates.
Portability changes how the product is built
Avoiding macros and external connections limits the mechanisms available to the workbook, but also reduces dependencies for the customer. Bounded formulas and explicit data structure carry the logic. A second office engine was used to open, recalculate, and save files to expose compatibility issues beyond the authoring environment.
Automated checks and visual review serve different purposes. Formula and package checks can identify structural problems, while rendered PDF inspection can reveal clipping, overlap, missing sections, or stray artifacts. Neither is treated as a substitute for the other.
Generated files still need an authorization model
The release process addresses four practical failure classes: broken workbook logic, cross-application incompatibility, accidental inclusion of internal files, and claims of success that have not been verified. The agent instructions put directory, account, credential, and release boundaries around the work rather than relying on an assistant to infer what is safe.
Customer packaging follows automated QA, visual review, and human approval. Release archives and package hashes support integrity checks, while internal QA artifacts remain outside the customer package. This applies least privilege, information minimization, and change control to an ordinary digital product.
Check both the workbook and the delivery
One historical Pet Care System delivery report records 1,358 formula cells checked and 49 QA checks: 48 passed, one non-blocking freeze-pane advisory, and no failures. The checks included formulas, workbook structure, package integrity, compatibility-related conditions, and customer-package contents. These numbers describe that artifact at that checkpoint, not every Clearward product.
Independent recalculation and rendered visual review were also reported. The recalculation checks the workbook through another engine; the rendered review checks what the customer will actually read. Storefront and listing coordination, customer packaging, and integrity checks complete the delivery work. No sales or adoption metric is inferred from those engineering results.
Turn an attractive template into a maintainable product
The move toward structured tables, stable identifiers, controlled inputs, and release checks demonstrates growth from document production into software-style product engineering. The workbook becomes something whose behavior can be reasoned about and tested, not just something that looks finished.
Directing the assistants through explicit permissions and independent verification also develops AI supervision. The useful skill is not only producing a file quickly; it is defining what a correct release looks like, detecting failures the generator may miss, and ensuring that the approved customer package is the one actually delivered.