{"resourceId":"st-petersburg-fl-ai-permit-review-limits-presubmission-20261009","versions":[{"version":"external-295ca936d73099ff76f3b46a5299b74ce73e19655528123a3b2fd596dd1c18df","resource":{"id":"st-petersburg-fl-ai-permit-review-limits-presubmission-20261009","title":"St. Petersburg, Florida tests 20+ AI plan-review tools, finds judgment and conflict limits, pivots to pre-submission assistance","organization":"City of St. Petersburg, Florida","sector":"Municipal building permitting and resident AI","geography":"St. Petersburg, Florida, United States","publishedAt":"October 9, 2026","publicationDate":"2026-10-09","eventDate":"2026-10-09","sourceName":"St Pete Catalyst","sourceLabel":"Independent local news report","sourceUrl":"https://stpetecatalyst.com/ai-could-change-how-st-pete-approves-building-plans/","evidenceClass":"independent-reporting","outcomeClass":"mixed","topics":["governance-procurement","operating-model","knowledge-work"],"finding":"After roughly six months and more than 20 vendor demonstrations, St. Petersburg technology staff told a council committee that AI plan-review tools struggled with judgment-heavy code (accessibility, door placement), conflicting rules, and high false-positive flag volumes. Special projects manager John Baker said AI is for now best suited for pre-submitted assistance. The city is implementing Tyler permitting software and Evolve completeness checks to cut missing-paperwork cycles, not to replace plan reviewers. Separately, a councilmember reported the city's website chatbot could not find a posted City Council agenda; staff cited content-currency difficulties. An August generative-AI policy already requires verification of AI output; public survey feedback showed more discomfort than comfort with AI.","sledRelevance":"Shows a mid-size U.S. city measuring plan-review AI against real plans and deliberately narrowing scope to intake completeness after documented limitations—directly on local permitting and resident access.","evidence":"Local reporting quoting city technology staff and councilmembers from a Public Services and Infrastructure Committee meeting; vendor demo counts and qualitative test findings; no published accuracy rates or cost figures.","architectureImplications":"Separate generative chatbots from dedicated plan-check tools; use AI for document completeness and applicant guidance before human code review; Tyler plus Evolve-style checks in the intake path.","governanceImplications":"August generative-AI policy requires verification; council debate on pace; public feedback more uncomfortable than comfortable; chatbot content ownership unresolved.","securityPrivacyImplications":"Policy restricts protected information in generative tools; plan-review pilots used actual construction plans with vendors—data-handling terms not detailed in the article.","caveats":"Secondary news account of staff remarks; quantitative accuracy, false-positive rates, and vendor identities for the 20+ demos not published. Corrected attribution of councilmember quotes noted in editor's note.","streamIds":["local-government"],"roles":{"sales":"Sales framing: St. Petersburg stress-tested plan-review AI on real plans and concluded it is not ready to decide code compliance. The credible conversation is about intake completeness, applicant guidance, and keeping reviewers accountable—not autonomous approvals. Discovery should ask which judgment-heavy code sections failed in demos and how false-positive volume was measured. Do not claim cycle-time savings; the city has not published them.","engineering":"Engineering framing: Fit is pre-submission document and completeness assistance integrated with Tyler, not autonomous plan approval. Architecture should log flagged issues for staff triage, keep human authority on conflicting code, and separate generative chat from dedicated plan tools. Proof of value needs false-positive rates on a golden plan set and a decline path when rules conflict.","delivery":"Delivery framing: Stand up intake prompts and completeness checks with named plan-review owners, then measure resubmission rates before expanding. Acceptance is fewer incomplete applications without shifting liability to software. Risks include chatbot content drift (already observed on agendas) and public distrust if marketed as automated review."},"roleDetails":{"sales":{"customerProblem":"Incomplete permit applications and pressure to use AI for faster plan review without proven code judgment.","stakeholders":["Building officials","Technology Services","City Council","Contractors and applicants"],"discoveryQuestions":["Which code provisions failed in vendor tests?","How will false positives be measured?","Who owns chatbot content currency?"],"valueHypothesis":"Pre-submission completeness checks may cut missing-paperwork loops; benefit not yet measured.","potentialEngagement":"Bounded intake-completeness pilot design with evaluation gates before any plan-decision automation.","unsupportedClaims":["Autonomous code compliance","Guaranteed permit cycle-time cuts","Chatbot accuracy"]},"engineering":{"fit":"Fits cities wanting AI for intake assistance while retaining human plan review.","architectureIntegration":"Tyler permitting workflow plus completeness checker; generative chatbot on separate content pipeline.","prerequisites":["Digitized application checklist","Golden plan test set","Content owners for resident chatbot"],"constraints":["Judgment-heavy code","Conflicting regulations","Florida public-records and liability norms"],"security":"Control vendor access to real plans; keep PII out of generative tools per city policy.","proofOfValue":["False-positive rate on held-out plans","Resubmission rate before/after intake prompts","Agenda/findability test for resident chatbot"]},"delivery":{"work":"Implement intake prompts and completeness checks; retain human reviewers; remediate chatbot content.","dependencies":["Tyler rollout","Evolve or equivalent","Department content owners"],"ownership":"Building/permitting with Technology Services support.","skillsAdoption":"Train reviewers to triage AI flags; train content owners for chatbot currency.","governanceCheckpoints":["Council AI discussion","Expansion beyond pre-submission assistance"],"acceptanceCriteria":["Documented false-positive metric","No unreviewed AI code decisions","Chatbot agenda findability fixed"],"risks":["Marketing overclaim","Content drift","Staff overload from noisy flags"]}},"retrievedAt":"2026-10-12T22:13:29-05:00","enrichedAt":"2026-10-12T22:17:23.042810-05:00","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Human reviewers retained for code decisions; incomplete chatbot and public discomfort signal accessibility and trust gaps for resident-facing AI.","procurementImplications":"Multiple vendor demonstrations without selecting a plan-review AI; Tyler permitting rollout expected spring; Evolve described for completeness checks.","operatingModelImplications":"Narrow use to pre-submission assistance; department directors propose uses evaluated for cost, security, performance; keep permitting AI under council review.","sourceVerification":{"openedUrl":"https://stpetecatalyst.com/ai-could-change-how-st-pete-approves-building-plans/","referenceExcerpt":"For now, Baker said, AI is \"best suited for pre-submitted assistance.\"","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}