{"schema_version":"2.0.0","id":"artifact:the-best-ai-system-should-need-less-ai-tomorrow","slug":"the-best-ai-system-should-need-less-ai-tomorrow","title":"The Best AI System Should Need Less AI Tomorrow","subtitle":"The Inference Retirement Thesis","thesis":"Use intelligence to conquer ambiguity, capture what works, and retire inference when uncertainty disappears.","abstract":"A flagship architecture thesis for AI-native systems: probabilistic intelligence should push into ambiguity, evidence should settle what worked, and repeated validated reasoning should migrate into deterministic machinery. Mature systems reserve inference for novelty, uncertainty, exceptions, and change instead of repeatedly paying models to rediscover solved procedures.","author":{"id":"person:tony-malott","name":"Tony Malott","url":"https://malott.ai/","resume_url":"https://malott.ai/resume/","email":"tony@malott.ai","email_url":"mailto:tony@malott.ai"},"type":"architecture-thesis-field-essay","featured":true,"public":true,"public_route":"/architecture/the-best-ai-system-should-need-less-ai-tomorrow/","public_safe_status":"reviewed","semantic_status":"locked","presentation_status":"locked","lifecycle":{"state":"PUBLISHED","owner_review":"accepted","merge_authority":"granted","production_authority":"granted","authority_axes":{"semantic":"complete","evidence":"complete","relationships":"complete","creative":"complete","render_uat":"qualified-pr-398-presentation-relocated-under-issue-403","owner_acceptance":"accepted"}},"dates":{"created":"2026-08-11","updated":"2026-08-11","published":"2026-08-11"},"topics":["ai-architecture","inference-retirement","deterministic-systems","agentic-systems","inference-economics","software-automation","control-planes"],"collections":[{"id":"collection:governed-ai-operations","title":"Governed AI operations","position":9.5}],"format":{"label":"Flagship architecture thesis and field essay","reading_time":"10 min","depth":"Cinematic visual thesis plus complete long-form article","interaction":"Three reader projections on one canonical route: visual brief, full thesis, and architecture dossier"},"card":{"kicker":"AI architecture · Inference economics · Deterministic systems","orientation":"Use intelligence to conquer ambiguity. Retire inference when the uncertainty is gone.","visual_motif":"frontier-cartography"},"presentation":{"motif":"frontier-cartography","visual_blocks":[]},"source":{"path":"content/artifacts/the-best-ai-system-should-need-less-ai-tomorrow/source.md","format":"markdown","content_status":"canonical-full","predecessor_sha":"a2767a4bb3ba8485e1913445a4cc649b65f9dc5c","public_projection_name":"Owner-approved Inference Retirement three-reader projection"},"source_authority":{"governing_issue":403,"prior_publication_pr":398,"qualified_pr_head":"7e4f319b3d3fc7992b2e695a24377d9a30a3cf84","prior_production_main":"074aaa64a988777c28b8ba3f6691b0deee93af98","predecessor_work":"artifact:inference-at-the-ambiguity-frontier","projection":"Owner-accepted Inference Retirement presentation relocated intact from the accidentally reused predecessor identity under Issue #403."},"evidence_authority":{"receipt_path":"docs/publication/issue-403/EVIDENCE_RELATIONSHIP_AUTHORITY.json","status":"complete","external_corroboration":true,"prior_art_disposition":"recorded","relationship_decision_ledger":"recorded"},"provenance":{"posture":"Owner-originated architecture thesis and first-person field synthesis previously published under the wrong predecessor Work identity by PR #398. Issue #403 corrects identity without rewriting the thesis or redesigning the accepted presentation.","ai_assistance":{"used":true,"role":"Collaborative thesis articulation, Tony Voice editing, visual architecture, source verification, evidence binding, and publication identity correction under direct owner authority."},"sources":[{"id":"source:shareplane-platform-issue-403","type":"governing-publication-authority","title":"Issue #403 identity split repair","locator":"https://github.com/pinklon/shareplane-platform/issues/403","role":"Governs the corrected independent identity and route.","publiclyExposed":true},{"id":"source:inference-retirement-pr-398","type":"prior-qualified-publication-lineage","title":"PR #398 — The Inference Retirement Thesis v2","locator":"https://github.com/pinklon/shareplane-platform/pull/398","role":"Preserves the qualified manuscript and presentation lineage.","publiclyExposed":true},{"id":"source:ambiguity-frontier-predecessor","type":"predecessor-work","title":"Inference at the Ambiguity Frontier","locator":"https://shareplane.malott.ai/architecture/inference-at-the-ambiguity-frontier/","role":"Foundational predecessor Work.","publiclyExposed":true},{"id":"source:shareplane-platform-pr-231","type":"internal-field-evidence","title":"Issue #230: build-once publication Fast Path v01","locator":"https://github.com/pinklon/shareplane-platform/pull/231","role":"Field evidence for deterministic build-once publication mechanics.","publiclyExposed":true},{"id":"source:google-adk-2","type":"primary-vendor-engineering","title":"Why we built ADK 2.0","locator":"https://developers.googleblog.com/why-we-built-adk-20/","role":"Corroborates deterministic application flow around probabilistic agents.","publiclyExposed":true},{"id":"source:openai-workspace-agents","type":"primary-vendor-guidance","title":"Workspace agents","locator":"https://openai.com/academy/workspace-agents/","role":"Corroborates deterministic workflow and probabilistic agent distinctions.","publiclyExposed":true},{"id":"source:anthropic-managed-agents","type":"primary-vendor-engineering","title":"Scaling Managed Agents: Decoupling the brain from the hands","locator":"https://www.anthropic.com/engineering/managed-agents","role":"Corroborates separation of reasoning from stable execution interfaces.","publiclyExposed":true},{"id":"source:openai-agents-transforming-work","type":"primary-vendor-empirical-context","title":"How agents are transforming work","locator":"https://openai.com/index/how-agents-are-transforming-work/","role":"Counterevidence that total agent use may still expand.","publiclyExposed":true},{"id":"source:compiled-ai-2604-05150","type":"research-preprint","title":"Compiled AI: Deterministic Code Generation for LLM-Based Workflow Automation","locator":"https://arxiv.org/abs/2604.05150","role":"Adjacent prior art for compiled deterministic workflows.","publiclyExposed":true}],"boundary":"An architectural thesis and first-person field synthesis, not a universal benchmark. External sources corroborate bounded mechanisms, while total organizational model use can rise as established subpaths retire inference.","research_gap":"Deterministic workflows and compiled-AI patterns have documented prior art. The lifecycle doctrine is presented as Tony Malott's synthesis without an originality claim."},"claims":[{"id":"claim:inference-retirement:primary","text":"Repeated validated inference should be intentionally retired from established execution paths when uncertainty has been resolved and successful behavior can be captured deterministically.","posture":"owner-architecture-thesis-with-adjacent-external-corroboration","support":["source:shareplane-platform-issue-403","source:google-adk-2","source:openai-workspace-agents","source:compiled-ai-2604-05150"],"caveat":"Some domains contain irreducible uncertainty and should retain inference."},{"id":"claim:inference-retirement:economics","text":"Repeated model-driven orchestration can impose recurring cost and latency where deterministic logic can execute the same stable path more cheaply.","posture":"externally-corroborated-bounded-systems-claim","support":["source:google-adk-2","source:compiled-ai-2604-05150"],"caveat":"Actual economics are workload-specific."},{"id":"claim:inference-retirement:maturity","text":"Mature architecture should create pressure to move sufficiently understood stable work from probabilistic reasoning into deterministic execution.","posture":"supported-synthesis-and-owner-lifecycle-thesis","support":["source:shareplane-platform-issue-403","source:google-adk-2","source:compiled-ai-2604-05150","source:openai-agents-transforming-work"],"caveat":"Total model use may still increase as agents move into new work."},{"id":"claim:inference-retirement:idr","text":"Inference Dependency Ratio is proposed as a maturity signal for established operational executions requiring model inference.","posture":"owner-proposed-metric-no-originality-claim","support":["source:shareplane-platform-issue-403"],"caveat":"Not standardized or empirically validated."}],"relationships":[{"type":"authored_by","target_id":"person:tony-malott","label":"Authored by Tony Malott"},{"type":"governed_by","target_id":"issue:shareplane-platform:403","label":"Governed by Issue #403"},{"type":"foundation","target_id":"artifact:inference-at-the-ambiguity-frontier","label":"Founded on Inference at the Ambiguity Frontier","explanation":"The predecessor Work establishes the durable boundary principle this Work develops into an explicit inference-retirement lifecycle."},{"type":"foundation","target_id":"artifact:the-code-is-no-longer-the-hard-part","label":"Start with the engineering cost inversion","reader_group":"foundations","explanation":"The cost-inversion thesis establishes the changing economics around AI-assisted implementation."},{"type":"foundation","target_id":"artifact:stop-prompting-agents-start-managing-workers","label":"Start with the governed worker model","reader_group":"foundations","explanation":"The worker-management thesis establishes roles, bounded authority, evidence, and earned autonomy."},{"type":"continuation","target_id":"artifact:the-agents-are-not-the-bottleneck-you-are","label":"Continue into the human-attention bottleneck","reader_group":"continue","explanation":"Once execution becomes cheap and deterministic, the remaining constraint moves toward human specification and review."},{"type":"companion","target_id":"artifact:frontier-inference-without-frontier-custody","label":"Companion: control the inference boundary","reader_group":"companions","explanation":"The custody guide governs what information may cross into frontier reasoning."},{"type":"companion","target_id":"artifact:the-most-expensive-model-is-often-compensation-for-missing-control","label":"Companion: reduce model dependence through control","reader_group":"companions","explanation":"Stronger intent, routing, validation, and control reduce compensatory model dependence."},{"type":"application","target_id":"artifact:governed-static-publishing-lane","label":"See the pattern applied to publishing","reader_group":"applications","explanation":"The governed publishing lane shows stable publication decisions becoming deterministic infrastructure."}]}