Frequently asked questions

Clear answers on the Six-Week Agentic Engineering Workflow Pilot, the Accountability and Evidence Map, the Initial Technical Review, supporting work for AI built into products, and what GhostCite does.

Working with Rule26

What does Rule26 AI do?

Rule26 provides client-side technical leadership for engineering leaders building and shipping AI in regulated and complex environments. The flagship engagement is the Six-Week Agentic Engineering Workflow Pilot, which improves one live engineering workflow for a team using coding agents. Rule26 also supports AI built into products, including architecture, evaluation and testing, and production readiness.

When should Rule26 be brought in?

When a team is using coding agents but its specification, review, testing and release practices have not caught up; when it is unclear who specifies, reviews, approves and accepts the risk of AI-assisted changes; when leaders need to see what reaches production; or when an AI feature is heading toward production and needs architecture, testing and production-readiness work.

Does Rule26 replace our internal technology team or implementation vendor?

No. We work alongside them while keeping a client-side, vendor-neutral view. Depending on scope we can lead a defined workstream, coordinate with vendors and internal teams, and provide temporary implementation leadership. We do not supply every specialty an engagement needs.

What does Rule26 not provide?

Legal, clinical, cybersecurity, insurance and regulatory advice, formal compliance determinations, certifications, and attestations. CIPP/US and AIGP inform the questions we ask; they do not make Rule26 a certifying authority. We examine privacy-relevant data movement and vendor boundaries, leaving formal conclusions to qualified professionals.

The Pilot and First Steps

What is the Six-Week Agentic Engineering Workflow Pilot?

Rule26’s flagship engagement. Rule26 works with one team using coding agents on one live engineering workflow, from request and specification through implementation, checks, client-owner review, release, and production follow-up. In weeks 1–2, we map the current workflow, identify decision points and risks, and design practical boundaries, checks, evidence, and named owners. In weeks 3–6, the team starts the pilot in its real work and we refine it as release cycles allow.

See the pilot →

What results should we expect from the six weeks?

An improved workflow started and refined in real work, with early pilot evidence of what changed. Six weeks may cover only a few release cycles, so the findings are early pilot evidence, not proven ROI or established long-term results. A separately scoped, paid 60-day recheck can assess later outcomes.

Who makes approval and release decisions during the pilot?

Your named owners. They retain approval, release, and residual-risk acceptance decisions throughout the pilot. Rule26 advises and supports the work; it does not make those decisions.

What is the Accountability and Evidence Map?

A lower-commitment first step for teams not yet ready for the pilot. Rule26 helps you document the responsibilities, decision points, checks, and evidence in one engineering workflow, with a named client owner for each decision. Your named owners retain approval, release, and residual-risk acceptance decisions. The map records who decides and what evidence they need. It is not assurance, and Rule26 does not independently assure a workflow it helped design.

What is the Initial Technical Review?

A limited $500 entry offer for one specific question, not equal in scope to the pilot. It covers up to 60 minutes working through one defined technical question or proposal with Salma Saad, followed by three prioritized risk and evidence observations and a concise executive follow-up document. Detailed solution design and implementation planning are completed as part of a paid engagement. It is an initial perspective, not a complete assessment.

See the Initial Technical Review →

AI Built Into Products

Does Rule26 still help with AI built into products?

Yes, as supporting work alongside the pilot or scoped separately: vendor and architecture review, AI evaluation and testing, productization and production readiness, and approved AI adoption. Scope is agreed for each engagement.

See supporting work on Services →

What does a vendor and architecture review examine?

Vendor claims and the evidence behind them, architecture and integrations, data and model boundaries, testing, logging and human review, ownership, dependencies and unresolved risks, and pilot or acceptance criteria where appropriate. The result supports your decision. It is not a contract review or a security certification.

What does AI evaluation and testing involve?

Depending on scope: defining intended uses and unacceptable failures, building or reviewing representative evaluation datasets, testing grounding, retrieval, permissions, exceptions and human review, classifying failures by severity, setting thresholds and regression tests, and assessing what the evidence does and does not demonstrate.

Is evaluation only for systems approaching release?

No. Evaluation can support a vendor pilot, an internal workflow, a release decision, a system already deployed, a significant change, an observed failure, or continuing regression testing.

How is evaluation different from productization?

Evaluation determines how the system behaves and what the evidence shows. Productization closes the engineering and operational gaps so the system can be integrated, operated, monitored, supported, and handed over responsibly.

What can productization and production-readiness work include?

Target architecture and integrations, authentication and authorization, logging, observability and auditability, evaluation and regression infrastructure, human-review and exception workflows, and deployment, rollback, escalation, ownership and handoff. Scope depends on the system and the specialist expertise required.

GhostCite

What does GhostCite verify?

GhostCite checks supported legal citations and quoted language against public court-record sources. It does not verify every legal proposition, provide good-law treatment, evaluate overall legal reasoning, or replace attorney review.

Open GhostCite →

Does GhostCite use AI for verification?

No. The verification step uses defined, rule-based checks against CourtListener public court-record data. No generative model decides whether a citation or quotation is supported. Optional semantic search is a clearly labeled fallback for locating a source, not a verification result.

Does GhostCite replace Shepardizing, KeyCite, or attorney supervision?

No. GhostCite confirms that a citation exists and that quoted language matches the record. It does not tell you whether a case is still good law, and it does not replace attorney review or professional judgment.

What happens to text submitted to GhostCite?

Submitted text is used to perform the check and is not used to train an AI model. Follow your organization’s data-handling rules when deciding what to paste into any external tool.

Still have questions?

Discuss Your Project