Archie, rebuilt.
A new reasoning engine behind every Archie conversation: an expert architecture team that reasons over the living model of your system, shows its work, and pushes back when the premise is wrong. Phase one of the rebuild: conversations now; blueprint generation and Archie in your coding assistant next.
What this means for you
The strategic calls are sharper.
The big calls, modernization, cost programs, consolidation, platform bets, now get what a top architecture team would conclude, worked over the whole of your estate's data rather than a sample of it. Reasoned over the living model, numbers that reproduce, a false premise refused: in a blinded ten-question comparison the new engine won all ten, with no fabricated facts.
Designs aligned to the system and the objective.
A spec your coding agents build from is now anchored in the live model and in your objectives, and a reviewer can check the path Archie took instead of re-deriving it. That improves the informed half of Mean Time to Informed Decision, not only the time.
Every conversation gets the same expertise, and leaves a record.
Day-to-day questions get the same engine and the same reasoning approach as the big calls. Every run leaves a record, and what you do with a blueprint is recorded when it happens: the record Archie will learn from next.
- Every Archie conversation in the console now runs on the new reasoning engine: an expert architecture team behind every answer. Cut over on September 18.
- Archie reasons over the living model of your estate, and a premise the estate contradicts is refused.
- It shows its work while it works, and always returns something you can act on.
- Its work is checked in ways suited to the task, and what you do with it is recorded.
An expert architecture team on every question
Archie now reasons the way a lead architecture team does. It picks the approach the question calls for and applies Catio's architecture reasoning: the right target for your objectives first, the gaps against it second, the highest-return moves ranked third. It changes its mind when your estate contradicts it.
Where the previous engine ran a fixed sequence of steps and wrote a confident paragraph at the end, the new one works the problem the way a team of senior architects would. Archie now carries Catio's architecture reasoning approach into everyday questions as well as full diagnostics.
Read more: Who is Archie?
Shows its work
Runs stream. Each step appears as it happens, described in plain language ("Reading the estate's cost position", "Checking what connects to this") rather than by what the tool is called. When the answer lands, the trail folds to one line above it, one click away. And Archie always returns something you can act on: when a question turns out to be bigger than the room it had, it gives its best answer and names the step it stopped at, so you know what to ask next. Never a silent timeout.
Trust that survives a second reader. An architect can audit the path in a click; a reviewer can approve the answer without redoing the work. Archie's live thinking streams while it thinks and is stored nowhere.
It pushes back
Given a premise the estate contradicts, Archie checks the live facts and refuses it. The measured example: told the migration off Kubernetes was finished and asked what to clean up next, the previous engine replied "Good news, your Kubernetes migration appears complete" and produced a decommissioning plan. The new one found the cluster active, its node groups modified that week and nine services still deployed on it, and declined to produce a cleanup list whose first item would take production down.
Across the same ten questions the previous engine contradicted itself four times, named a retired model as current spend, and declared a running cluster empty. The new one did none of those things. A blinded judge, shown both answers without knowing which engine wrote which, preferred the new engine on all ten, with the widest gap on technical merit. One caveat, stated: ten questions, one workspace, one blinded pass. A direction, not a rate.
Checked before it reaches you, recorded after
Before it answers, Archie applies checks suited to the task: blueprint outputs are evaluated, while the checks on a chat answer vary with the question. Numbers that matter, a cost breakdown, a reconciliation, are computed rather than reasoned about, so twelve repeats give the same checked total. And every run leaves a record: what you do with a blueprint, view it, comment, accept or decline it, move it along, is recorded when it happens.
A check that always approves is not a check. A business case whose totals a second reviewer cannot reproduce is not a business case. And what you do with a recommendation is the one signal its score can ever be validated against, so it is recorded now. Learning from it is next.
Read more: Blueprints
What we measured, and what we did not
The same engine, everywhere
- Blueprint generation on the new engine shipped on October 5, for every customer; the October post covers it.
- Archie in your coding assistant, over MCP, on the new engine.
- Archie learning from its own run history, with a person approving what changes.
Elsewhere in the console this month
- Sep 24Blueprint status lifecycle: Draft to Completed pills, a tracker and plan progress; bookmarks retired.
- Sep 24Blueprint diagrams read relationship subtypes from the component library and draw edge labels above grouped edges.
- Sep 23Blueprints: multi-select with bulk delete and add-to-plan, plus a Creator column.
- Sep 22Stacks show what a component does, not only what it is.
- Sep 21Workspace Context: the Sources list pages, and a failed indexing run no longer strands a document.
- Sep 16Stacks render faster.
- Sep 15Processes: cost report runs are rendered by name; workspace access lists name their principals.
- Sep 08Plans link to the source files they were generated from.
The console, this month
Two minutes on the changes above, recorded in the console: where every blueprint stands, the review path, reactions, plans, diagrams you can hand to someone, version history, Stacks, and a chat with Archie.
Want to see it on your own system? Book a demo