Pimp My IDE / Garage Dispatch
← Back to garage
September 16, 2026 · model routing / typed decisions / live agents

Stop sending every decision through a chat box.

One new model throws away prose to return typed probabilities. Another keeps talking while tools run. A hot review harness bolts deterministic rails around a language model. The signal is not that one architecture wins. It is that “AI” is finally splitting into drivetrains.

THE TAKE: route by output shape. Use language for exploration, typed probabilities for bounded judgment, deterministic code for invariants, and a human brake when a confident mistake can hurt somebody.
Tune the Decision Shape Manifold ↓

A string is not the natural shape of every answer.

TypeSafe AI’s early-access Jev model accepts unstructured state and returns predefined typed values with probabilities rather than generated prose. The company positions that design for classification, routing, scoring, extraction, branching, and large map-reduce jobs; it reports 70–500 ms service latency and large cost and speed advantages on its own workflow evaluations.[1]

The interesting part is the contract, not the drag-strip number. The benchmarks, reference answers, task selection, and “can’t hallucinate” framing come from the vendor. A value can be type-correct, well-calibrated in aggregate, and still wrong on the case that matters. Typed output removes one failure class—malformed freedom. It does not remove model error, bad decomposition, drift, or the cost of a wrong branch.

Do not ask prose to impersonate a control signal.

Some jobs need conversation while the wrench is moving.

Google’s Gemini 3.8 Live launch takes the opposite shape: voice, visual context, interruption handling, background tool calls, and an Extended Thinking mode that speaks progress cues while multi-step work continues. Google reports benchmark results and support for automatic switching across 97 languages; those are vendor-reported launch claims, not measurements from this garage.[2]

That architecture makes sense when ambiguity is the work: troubleshooting, gathering constraints, coordinating a booking, or shaping a UI through spoken feedback. Rich language is not waste there. It is the interface. But narration should not become authority. “I’m checking that” is feedback; it is not a transaction receipt.

The harness gets the boring guarantees.

Alibaba’s Open Code Review makes the third move. Deterministic code chooses files, bundles related changes, matches rules, and positions comments; an agent searches context and exercises judgment inside those rails. The project deliberately describes lower recall in exchange for higher precision, and its scale and benchmark numbers are maintainer claims.[3]

That split is the shop manual. Put enumeration, schema validation, budgets, permissions, retries, idempotency, and audit records in ordinary code. Put fuzzy classification in a bounded decision model when the option set is known. Put exploration, explanation, and synthesis in a language model. Put irreversible, high-consequence decisions behind a named human.

Design the abstain lane before the fast lane.

A useful router needs more than a model picker. Define the allowed outputs. Set the consequence of a wrong answer. Capture confidence, but test calibration on your traffic. Specify an abstain or escalation threshold. Log input version, model version, decision, probability, downstream action, and later disposition. Then replay changed models against the same packet before switching the route.

The manifold below replaces the universal “send to AI” button with three visible routes and three interlocks. Its gauge is a teaching proxy for declared automation pressure, not production telemetry or model quality.

Interactive makeover / routing bench

Decision Shape Manifold.

Traditional purpose replaced: one “use AI” toggle. Better version: choose the output contract, set consequence, close calibration and escape interlocks, then print a route receipt. Native radios, checkboxes, and range input preserve keyboard and screen-reader operation.

Route the workload

Pick the result shape before picking the model. The gauge shows declared automation pressure only.

Output contract
2 / 5
Automation interlocks

Declared automation pressure

CONFIG / 100 · TEACHING PROXY

PROSE lane at consequence 2/5. 0/3 interlocks declared. Use language for exploration; require a separate action boundary.

Open the three-source evidence bay
[1] TypeSafe AI, “Introducing System One Models & Jev,” September 15, 2026 — architecture, intended workloads, pricing, latency, workflow evaluations, and the vendor’s own caveats. [2] Google, “Introducing Gemini 3.8 Live and 3.8 Live Extended Thinking,” September 15, 2026 — live dialogue, visual grounding, language switching, background tools, progress narration, and vendor benchmarks. [3] Alibaba Open Code Review repository and README, read September 16, 2026 — deterministic/agent split, review surfaces, precision/recall tradeoff, and maintainer-reported deployment and benchmark claims.

Source boundary: this dispatch compares published system shapes. It does not independently benchmark the models, certify calibration, or endorse a vendor. The manifold records declared controls; it does not execute a workflow or prove a route safe.