HR AI moves from describing candidates to judging them, with the scoring rubric handed to the customer.

The week in hr-recruiting

The clearest line through the sector this week is AI moving from describing candidates to judging them, and vendors racing to put the controls for that judgment in the customer's hands. HiBob replaced its generative CV summary with a configurable, explainable match score built on recruiter-defined criteria; Spark Hire pushed its AI past pre-interview screening and into the interview room, where an AI Notetaker now drafts the pros, concerns and scorecard ratings a hiring team signs off on; and Workable opened up the Ideal Candidate Profile its screening agent runs on, letting recruiters reword criteria and re-score a pipeline against the revised rubric. Three different products, one shared instinct: automate the evaluation, but hand the customer the rubric and the audit surface that makes an automated decision defensible.

The second move is suite expansion — products outgrowing the category they were bought for. Workable keeps shipping like an HR platform rather than an applicant tracker, Ever Gauzy bolted a full document hub onto its open-source ERP, and two products stepped outside HR entirely: Envoy turned its workplace record into a threat-monitoring safety layer, and Workyard let its AI time assistant start moving job costs between projects. Compliance plumbing rounds out the week, with Tanda emitting UK statutory pension files from a payroll engine built for Australian awards.

Leaders

Ever Gauzy had the busiest changelog in the sector — ten release tags on 8 August alone — anchored by a single spark: a Documents hub with wiki pages, sharing, quotas, a review workflow and an AI knowledge pipeline, shipped alongside a full swap of CKEditor 4 for a TipTap editor across the suite. Most of the surrounding tags exist only to service that one merge, so its version numbers run well ahead of its actual change volume.

Tanda shipped the week's clearest compliance signal: PAPDIS 1.1-compliant pension export files generated straight from a completed pay run, with validation surfaced in the flow, for UK pension providers. That is Tanda's award engine emitting statutory output well outside its Australian roots, and it arrived amid a steady run of rostering and reporting refinements.

HiBob swapped its AI CV Summary for AI CV Matching — an explainable match score against criteria the recruiter defines, rather than a generated paragraph. It moves the AI from describing candidates to ranking them, which HiBob itself frames as a different product and a different liability.

Spark Hire crossed the same line from the interview side. Its new AI Notetaker sits in on interviews, transcribes them, and then proposes the evaluation itself — pros, concerns, areas to review and scorecard ratings — which teammates accept, edit or skip before submitting. Every prior AI release scored or drafted inputs; this one drafts the assessment a hiring team signs off on.

Workable made its screening agent's rubric editable. Recruiters can now reword a criterion, add or drop one, change what counts as a disqualifier, and re-evaluate the pipeline against the revision — free for recently created stages, metered beyond. It moves the agent from a fixed evaluator set at job creation to a rubric the customer owns and revises.

Wildcards

Envoy shipped three releases at once under a new Response line, led by Threat Intelligence: a dashboard that continuously monitors verified global and local risks against the locations a customer already has on file, extended to employees' home addresses once an HR directory is synced. It is a sharp break from visitor sign-in and desk booking toward duty-of-care safety, built on the workplace record Envoy already holds.

Workyard let its AI Time Assistant graduate from correcting clock times to reallocating hours between projects and cost codes from a chat prompt, showing the old allocation struck through beside the new one before anything commits. Natural-language job costing is an unusual place for a construction workforce tool to land, and it completes the pre-payroll cleanup loop the assistant has been building toward.

Themes that compounded

• Explainable, customer-owned screening rubrics: HiBob, Spark Hire and Workable all shipped AI that produces a judgment while handing the criteria and audit surface to the customer.
• AI crossing from inputs to assessments: the newest releases from Spark Hire and HiBob draft or score the evaluation itself, not just the data behind it.
• Suite expansion past the core category: Workable into HR, Ever Gauzy into a document hub, Envoy into workplace safety.
• Compliance and integration plumbing as product: Tanda's PAPDIS exports and Workyard's cost-code reallocation both remove a manual re-keying step ahead of payroll.
• Release churn masking change volume: Ever Gauzy's one-merge-many-tags pattern inflates its cadence well past its real work.

Watch this week

Watch whether the screening-AI leaders add the audit trail their new controls imply: Workable's editable Ideal Candidate Profile and HiBob's configurable match score both invite a demand for versioned criteria and a record of which rubric scored which candidate. On the expansion front, Envoy's Response line describes monitoring and mapping but not escalation, so alerting and follow-through workflows are the obvious next addition, while Ever Gauzy's Documents churn should settle as the feature clears stage — and Tanda's move into statutory output beyond Australian awards is worth tracking for more formats at the same cadence.