AI systems that survive contact with production.
The unit of change is a workflow, not a chatbot.
A chatbot stops at an answer. A workflow completes the job. Delightful Computer redesigns measurable business processes—intake, review, research, reporting—and embeds AI into the systems and steps the team already uses. The result is not generic AI enablement or slide-deck strategy. It is production software that can survive procurement, compliance, subject-matter expert review, and real operator use.
The practice is deliberately small and direct. It maps a high-value workflow, baselines its operating cost and delay, identifies the data and decision boundaries, builds a production slice, instruments it, and transfers the code, runbook, evaluation suite, and operating notes to the client team. Typical projects emphasize retrieval, model routing, agent orchestration, evaluation, observability, secure deployment, and human-in-the-loop review.
Who Runs It
Delightful Computer is Morgan, working solo. Engineering sits on the critical path from workflow map to deployed system: the same person writes the code, sits with the subject-matter experts, and owns the outcome. Roughly a decade designing production software for regulated enterprises, the last several years on LLM-native systems. Morgan also works on agentic AI at Microsoft; Delightful Computer continues as an independent practice with a deliberately small client roster and engagements scoped to stay conflict-free.
The domain matters less than the shape of the problem. Good-fit work has a measurable operational bottleneck, subject-matter experts who know the job better than any model, constrained data access, a human reviewer with real authority, and a clear definition of what production means. In practice that has meant regulated and high-trust settings most often — but the constraint set, not the vertical, is what makes the work tractable.
What Delightful Computer Builds
The primary output is production software, not a proof of concept. Systems usually combine private data access, retrieval pipelines, structured model outputs, workflow-specific user interfaces, reviewer queues, telemetry, and operational runbooks. The implementation is chosen around the client environment: Azure OpenAI, AWS Bedrock, existing identity providers, existing document repositories, existing ERP or LIMS surfaces, and the security controls already in force.
The work starts with workflow mapping because most failed AI projects choose the wrong primitive. A useful system needs the task boundary, exception path, review point, evidence requirement, and rollback path before anyone argues about prompts. Delightful Computer therefore treats evaluation, observability, and audit trails as part of the product surface from day one.
Services
Discovery is a two-week workflow map, data-access plan, architecture spike, success metrics, and make-buy-partner recommendation. Production Spike engagements ship one workflow end to end with an evaluation suite, guardrails, observability, alerting, runbook transfer, and a 30-day care window. Partnership is a sustained-operation retainer for a system already in production: continuous eval monitoring with alert thresholds, ownership of the model migration path, a bounded monthly change budget, named-hours incident response, and a quarterly written report for leadership. It is bought at handover after a Production Spike, or cold by a client who inherited an AI system from another vendor and has nobody on staff who can maintain it.
Engagements are fixed-scope and fixed-fee where possible, quoted against one named workflow rather than a public rate card. The goal is to make the edge explicit: what workflow ships, what data it can touch, who reviews outputs, what the model is allowed to decide, what it logs, and what happens when it is wrong.
Selected Production Work
Public case studies include a regulated-lab audit intelligence system for a pharmaceutical environment, a multi-agent legal research copilot, an instant sales brief generator for a financial-services consulting workflow, quote and intake automation for home-services operations, and law-firm prototype work. These pages describe the business problem, technical approach, implementation tradeoffs, results, and lessons learned.
Security and Privacy Posture
Delightful Computer designs for client-owned deployment, scoped credentials, least-data access, reviewer-visible reasoning, logging, and rollback. The public website does not expose customer workspaces, operational write actions, or a public production API. Client data handling is defined per engagement through the statement of work, data-access plan, security review, and target architecture.
Security reports can be sent to morgan@delightful.computer. Public security and privacy notes are available at Security and Privacy.
Developer and Agent-Readable Resources
The site publishes public machine-readable files for crawlers, answer engines, and agent tools that need a concise description of the company, services, constraints, and canonical pages. These files are informational only. They do not grant access to customer systems, private data, or write actions.
Writing and Field Notes
The writing section covers agent architecture, reliable generative AI systems, evaluation-driven development, RAG quality loops, telemetry envelopes, cache-aware prompts, typed structured outputs, workflow routers, and the difference between API wrappers and durable agent operating systems. These essays are intended for technical buyers and builders who need concrete implementation language rather than generic AI commentary.
Read writing, open the Pattern Language, or open the RAG Quality Guide.
Contact
To evaluate a project, send a note describing the workflow, stakeholders, existing systems, data constraints, review requirements, and what done would mean in production. You can also book an intro call. The fastest useful conversation usually starts with one narrow workflow and a concrete failure mode: slow review, inconsistent research, missing citations, manual intake, brittle reporting, or a runbook nobody trusts.