Products · speaker script
00:00–00:45 · SLIDE 35
Meet the Lintel family.
SAY Four packages, with a different job for each.
DO Pause on the four motifs. Name each product once. Click an image to open the standalone artwork; each product label jumps to its slide.
THEN Start with the complete system.
Four products, four jobs. A portal for the complete system, layers for team context, an open frame for a focused workflow, and a reticle for measurement. Same family, different responsibilities. Think Unix philosophy, with slightly better lighting.
Lintel is the complete harness. The CAIP Pack brings our role and team context. Core is the planned smaller shell for a specific customer case. Benchmark is the planned way to test whether the added process is worth it.
They have different release states. Lintel is available in the public repository. The CAIP Pack distribution link will follow. Core and Benchmark are planned offerings; we will add their links later. The prepared comparison in this presentation is already available, but it is not the complete Benchmark product.
00:45–02:15 · SLIDE 36
One task. A useful first session.
SAY Start with an accepted change, not a tour of every command.
DO Walk select → plan → verify → resume. Link to the getting-started guide.
THEN The pack supplies the organisation-specific starting point.
Choose a reviewed revision and exact client adapter. Use the manual START file if native discovery is unavailable. Plan a bounded health endpoint with an acceptance test. Inspect conventions, authorize the plan, execute the package and obtain separate review.
Then start a fresh session and verify the same work and next action. This is the first useful installation test: an accepted result and usable handoff. The current guide has the complete walkthrough. Installed files alone do not prove host discovery or execution.
02:15–03:15 · SLIDE 37
Your role. Already in the room.
SAY The CAIP Pack brings team context to the universal method.
DO Read the workshop request, then trace the added context to a more relevant brief. Describe this as the pack’s intended use.
THEN What if a customer only needs one focused workflow?
Imagine the request: prepare a customer technical workshop. Without the relevant team context, the agent has to rediscover the audience, expected evidence and our way of working.
The purpose of Lintel CAIP Pack, Pack number one, is to carry the MS employee in CAIP role and the approved team knowledge and standards that belong with it. The universal method stays in Lintel. The team-specific content stays in the pack. The example on this slide shows the intended value, not a live execution of a released pack.
For a solution engineer or architect, this could mean a workshop brief that starts with the right audience, a useful technical agenda and explicit questions still to resolve. The operator remains responsible for the customer context and approved inputs.
The pack’s distribution link will follow. This public presentation contains no private employee pack.
03:15–04:45 · SLIDE 38
One customer case. A smaller harness.
SAY Core is the planned shell for a focused workflow inside the customer’s coding app.
DO Trace the explicit start, the chosen workflow and the reviewed artifact. Use architecture review as the example.
THEN A smaller workflow still needs to prove that it helps.
Core is the planned stripped-down version of Lintel. Start with the customer’s case and select only the context, workflow, records and checks it actually needs.
The intended interaction is explicit: a person asks to run the architecture-review workflow in the coding app they already use, including desktop GitHub Copilot. The harness helps produce a review and decision brief, with enough state to return to the work. It does not need to invoke itself automatically. That is the design intent for Core, not a claim that every host or existing full Lintel install already behaves this way.
This can avoid maintaining a separate application or hosted agent flow for a bounded, person-driven task. It still takes ownership, versioning and testing. It is not zero maintenance.
When a case needs unattended execution, shared services, durable jobs or enforced platform controls, hosted infrastructure still has a job. The question is which runtime the use case requires. Core’s distribution link will follow.
04:45–06:00 · SLIDE 39
Does the harness earn its place?
SAY Benchmark should measure accepted work, including the cost of the harness.
DO Point at what stays the same and what changes. Name the four measures. Open the prepared comparison only if time allows.
THEN Choose one recurring task and measure the result.
Lintel Benchmark is the evaluation direction. The current kitchen harness now defines three arms: Naked Copilot, Lintel serial and Lintel + Swarming. All share the Service House brief and 32 requirements for 100 points. It records usable UI, declared completion, requirement coverage and billed credits separately.
The harness and reference/control checks exist locally. The public page describes the method; it does not bundle the local server, publish trial/session records or claim completed comparative runs. The distribution link remains pending. This is not evidence of a productivity win. Normal Lintel versus naked also changes delegation/model mix; keep the caveat visible and repeat with matched models if isolating that effect.