On May 15, 2026, GitHub Blog published Building a general-purpose accessibility agent — and what we learned in the process. The general-purpose accessibility AI agent built by GitHub's team provides an end-to-end workflow by scanning live web pages to handle WCAG 2.2 auditing → remediation proposals → automated PR generation. The threshold of a new era where "agents handle audits that humans can no longer keep up with" has finally become operational at an industrial scale.
For teams managing mid-sized enterprise websites over the long term, this serves as "a key differentiator for 2026–2027." In Japan, revisions to the Act on the Elimination of Discrimination against Persons with Disabilities made reasonable accommodation mandatory for private enterprises starting in 2024, expanding the scope of responsibility in custom development as covered in our Web Accessibility Guide. This article outlines the phase structure, technology stack, and contract clauses required when agencies design "WCAG audits + managed remediation operations" as an ongoing service.
Why manual WCAG audits have reached their limit
| Challenge | Manual audit | AI agent audit |
|---|---|---|
| Audit frequency | Every 6 to 12 months | Daily / per commit |
| Audit scope | Key 10–30 pages | All pages + dynamic states |
| Detectable items | Primarily visual and structural | + Contrast / focus transitions / ARIA |
| Remediation proposals | Written report only | Automated generation up to fix PRs |
| Executive visibility | Snapshot at audit time | Daily trends + improvement rate |
| Effort | 80–200 hours per audit | 2–4 hours per cycle |
This marks a shift from a model of "hiring an auditing firm once every six months" to one where "the custom development firm runs continuous operations." It serves as a powerful weapon for lowering churn on ongoing client development retainers.
Three structural shifts driven by accessibility agents
Shift 1: Moving from "auditing" to "guardrails"
Because automated audits run on every PR, accessibility violations are blocked before reaching production. This is a classic application of the harness design for integrating AI agents into operations discussed in Introduction to Harness Engineering.
Shift 2: Moving from "page-level" to "journey-level"
Agents audit by reproducing user journeys. You can guarantee WCAG compliance at the business scenario level, such as "through contact form submission" or "through checkout completion."
Shift 3: Remediation costs approach virtually zero
GitHub's agent automatically generates fix PRs. Agency engineers can focus their effort on "PR reviews and merge decisions," dramatically boosting price competitiveness.
Five phases for designing client "WCAG audits + managed remediation operations"
Phase 1: Current-state audit + conformance level selection (2 weeks)
Scan the client's entire site once with the AI agent and agree with the client on the WCAG 2.2 conformance level (A / AA / AAA). Set targets according to legal requirements, user personas, and industry standards.
Phase 2: Audit pipeline construction (3–4 weeks)
Integrate the AI accessibility agent into CI. Automated audits run on every pull request, treating threshold violations as CI failures.
Phase 3: Remediation backlog curation (4–6 weeks)
Map violations detected during the initial scan onto a "severity × remediation cost" matrix, breaking them down into a 3- to 6-month sprint plan. Give top priority to revenue touchpoints like checkout and contact forms.
Phase 4: Monthly reporting + executive meetings (ongoing)
Report monthly on "WCAG conformance rate / violation count trends / resolved PR count / journey coverage" to client leadership. CC the legal department so the reports can serve as evidence of meeting reasonable accommodation obligations.
Phase 5: Legal coordination + audit trail preservation (ongoing)
Preserve audit logs, remediation PRs, and executive minutes in 5-year increments, establishing them as an evidence trail for potential litigation and complaint handling.
Standard technology stack set for custom development
| Layer | Recommended technology | Alternative |
|---|---|---|
| AI agent | GitHub Accessibility Agent | axe-core + in-house Claude / Codex integration |
| Static auditing | axe-core / pa11y | WAVE |
| Dynamic scenario auditing | Playwright + axe-core | Cypress |
| CI integration | GitHub Actions | CircleCI |
| Dashboard | Looker Studio + BigQuery | Datadog |
| Automated remediation PRs | Claude Code / Codex Agent | GitHub Copilot Agent |
| Evidence preservation | GitHub + Notion | Confluence |
Executive visibility can be enhanced particularly through Notion integration, as covered in Custom Development with Notion Developer Platform.
Which projects need this and which do not
| Projects requiring this | Projects not requiring this |
|---|---|
| B2B / B2C sites with 50K+ monthly PV | Internal-only tools |
| Sites with checkout / application / registration forms | Static landing pages only (short-lived) |
| Transactions in public sector, finance, or healthcare | Limited industries |
| Operating a recruitment site | Not actively hiring |
| Legal department prioritizes reasonable accommodation | No legal framework established |
Six clauses to include in client contracts
| Clause | Details | What the client should verify |
|---|---|---|
| Target conformance level | A / AA / AAA | Legal requirements |
| Target page scope | All pages or primary user flows | Liability outside scope |
| Remediation PR approval authority | Need for client approval | Permission for auto-merging |
| Evidence retention period | Retention of audit logs and PRs | 5 years / 7 years |
| Legal coordination | Response to litigation and complaints | Coordination with legal counsel |
| Conformance rate SLO | Monthly WCAG conformance rate target | Measures upon non-attainment |
Estimated client-side ROI (assuming a mid-sized company of 300 employees with hiring and service sites)
| Item | Without audit | With audit | Difference |
|---|---|---|---|
| WCAG 2.2 AA conformance rate | 35% | 92% | +57pt |
| Complaint handling effort | 40 hours/month | 5 hours/month | -35h |
| Application conversion rate (including diverse demographics) | Threshold | +15〜30% | — |
| Litigation risk coefficient | High | Low | — |
| Legal audit compliance cost | 2 million yen/year | 600,000 yen/year | -1.4 million yen |
| Annual benefit | — | — | Approx. 8 million–15 million yen |
The savings from complaint handling and legal audit effort alone are substantial, putting first-year ROI recovery well within reach.
Five common pitfalls
Pitfall 1: Leaving goals at "targeting AA" without follow-through
As the legal landscape shifts, there are industries where even AA is insufficient. Continuously agree with the client on industry-specific baselines.
Pitfall 2: Client engineers unable to review remediation PRs
Reporting only PR counts breeds distrust when leadership and legal want to see that "issues are being addressed." Make progress visible with dashboards.
Pitfall 3: Skipping physical screen reader testing
AI agents focus primarily on code analysis, whereas real screen reader speech behavior is entirely different. Include monthly physical device testing in the contract.
Pitfall 4: Excluding dynamic content (modals and toasts) from audit scope
Dynamic UIs are breeding grounds for accessibility violations. Always combine audits with Playwright scenarios.
Pitfall 5: Prioritizing fixes purely from a technical standpoint
Disconnecting fixes from business impact results in merely "fixing the easiest bugs first." Align priorities with marketing and legal.
90-day action plan
| Week | Action |
|---|---|
| Week 1〜2 | Full site scan + conformance level agreement |
| Week 3〜6 | CI integration + AI agent operationalization |
| Week 7〜10 | Backlog burndown (15 PRs/month) |
| Week 11〜13 | Monthly meeting launch + legal coordination |
Summary — The era of continuously managing reasonable accommodation obligations for clients
The release of GitHub's general-purpose accessibility agent transforms WCAG operations from "manual auditing and fixing" into a "continuous, automated" process. For agencies, this is the strongest proposal asset for "mid-sized companies seeking to outsource legal risks."
Whether you are facing situations where "two years have passed since the last accessibility audit," "it is unclear if you comply with revised laws," or "diversity compliance on your recruitment site has fallen behind," we provide custom estimates based on site scale and target conformance levels. Please feel free to reach out via our contact form.
