Responsible Signatory Load Balancer for Perth Boutique Firms: Keep Partner-Level Review Honest When Juniors Are Quietly Using AI
You run a seven-lawyer firm in West Perth. Two partners sign almost everything that leaves the door. You suspect — without proof — that at least three people downstream are pasting drafts into a consumer chatbot to “tighten things up”. The partner whose name is on the cover page is the one carrying the candour-to-the-court obligation, and the partner who happens to be free at 4:45pm on a Thursday is the one who signs. That’s not a review system; that’s a queue. The Responsible Signatory Load Balancer is built to make that queue legible, and to make sure no single partner is asked to sign more AI-assisted work than they can actually verify.
Why it matters now
Shadow AI use inside a small firm is rarely malicious — it’s productivity pressure, plus the absence of a tool the firm has formally sanctioned. The problem is that the partner signing the document inherits responsibility for any AI-generated content in it, whether or not they knew a model was involved. For matters before the Administrative Review Tribunal, the ART’s expert evidence guidance places obligations on practitioners around the integrity and provenance of expert material — obligations that are difficult to discharge if the partner signing has no visibility into which parts of a submission, brief, or instructing letter passed through a generative tool. In a boutique firm, signatory capacity is the binding constraint. If one partner is doing 70% of the sign-offs because the other is in hearings, the review is functionally rubber-stamping. That is the failure mode this agent is designed to interrupt.
The 5-minute view
- “Shadow AI” describes the use of generative AI tools by staff without firm-level sanction, logging, or partner visibility — most commonly via personal accounts on consumer chatbots
- In firms with only two or three signing partners, signatory capacity becomes the bottleneck and the de facto cap on quality review
- Responsible signatory obligations under the Australian Solicitors’ Conduct Rules and the ART’s expert evidence guidance attach to the partner whose name appears on the document, regardless of how the document was drafted
- A load balancer doesn’t replace review — it makes sure review demand is matched to review capacity, and that no partner is silently absorbing more AI-touched output than they can verify
- The agent tracks which documents have been flagged as AI-assisted, which partner is on the roster for that matter type, and current queue depth per partner
- Output is a daily allocation view: who’s signing what, what’s waiting, and where the queue is about to break
- Runs on firm-controlled infrastructure; no draft content leaves the firm
What Exegesis is building
The Responsible Signatory Load Balancer Agent is a tier-3 service in the Exegesis Legal stack. It sits between document production and partner sign-off. When a draft is submitted for signatory review, the agent records the matter, the document type, whether AI assistance was declared (and by whom), and the current queue depth for each eligible signing partner. It then proposes an allocation — not to randomise partner exposure, but to keep AI-assisted volume per partner inside a configurable threshold the firm sets for itself. Where a queue is approaching capacity, the agent surfaces it before a 4:45pm rush forms. The architecture is deliberate: the agent reasons about queue state and allocation, not about the content of the drafts themselves. Drafts stay on firm infrastructure. The agent’s job is to make the workload visible, not to read the work.
Verification of citations and authority references inside those drafts is handled separately by RuleCheck, the open-source pre-lodgement checker, which can run alongside the load balancer in a typical boutique-firm workflow.
The deliverable
- A live signatory allocation view per partner: current queue, AI-assisted volume this week, next document due
- Per-document metadata capture: matter ID, document type, declared AI assistance, originating fee-earner
- Configurable per-partner thresholds for AI-assisted documents reviewed per day/week
- Alerts when a partner’s queue is approaching the threshold the firm has set
- Weekly summary report suitable for partnership meetings: where signatory load is concentrating, where shadow AI is being declared (and where it likely isn’t)
- Audit log entries timestamped per allocation decision, retained for the firm’s configured period
CTA
This is a tier-3 service shape and we’re scoping it with a small number of design partners before broader release. We want input from firms with two-to-five signing partners on what the allocation thresholds should default to, what fits inside a partnership meeting reporting cadence, and how the agent should handle expert evidence matters specifically. Join the waitlist and we’ll bring you into that conversation.
Sources
- Administrative Review Tribunal — Practice Directions and Other Guidance: https://www.art.gov.au/help-and-resources/professionals-and-practitioners/practice-directions-and-other-guidance
Exegesis capability references: