branco branco
Sign Up
Book a Demo
How Branco's AI Agents Work — Branco
AI Commitments

How Branco's AI agents work and what they will never decide for you

A signed, dated commitment by Stefano Grossi, CEO.

Last updated 19 May 2026

Is Branco safe to use?

Yes — and the reason is structural, not aspirational. Branco's three AI agents help your team give and receive feedback that moves work forward. They do not rate humans, set compensation, or decide who gets hired, fired, or promoted. Those decisions stay with named humans, by written commitment.

Branco runs three AI agents — Feedback Gatherer, Feedback Scorer, and Action Planner — on the same product stack as our performance management cycle. They nudge people to give feedback on time, score whether feedback is specific enough to be useful, and draft growth plans for a human to edit.

They do not run conflict resolution or write your performance review.

If you turn the agents off, the platform still works. The cycle, calibration, reviews, and action plans all run without AI. You lose speed and a quality signal. You do not lose the platform, and you do not lose your data.

This page covers what each agent does, what it sees and does not see, what we will never let it decide, and what shaped every commitment below.

What Branco's AI agents do

Three agents, three jobs, all narrow.

Feedback Gatherer

Initiates weekly feedback prompts. Pairs peers. Sends the 3–5 question prompt through Slack or Teams. Routes the response back to the right OKR, skill, or competency tag.

It does not write the feedback. It does not interpret the feedback. It moves a prompt into the right inbox at the right time and routes the reply.

Bounded to: prompt delivery and response routing.

Feedback Scorer

Reads feedback text and assigns a feedback_quality_level score — whether the feedback is specific, dated, and tied to an observable behaviour or outcome. The score is visible to the feedback giver as a coaching nudge and to managers as a calibration aid.

It scores the signal, not the person. It is not a rating of the recipient.

Bounded to: feedback text quality assessment.

Action Planner

Drafts a personalised growth plan from a body of feedback. Suggests Small Wins — micro-milestones with target dates. Every step is a suggestion. Every Small Win has an Accept or Reject button.

The human owns the plan. The agent drafts the proposal.

Bounded to: plan drafting and suggestion.

If you are looking for an agent that writes performance reviews, sets ratings, recommends comp, or auto-files PIPs — Branco does not ship one. The next section says why.

What Branco's AI agents will never decide for you

There is a hard line under every commitment on this page. Branco's AI agents will not — by product design and by written policy — make any of the following decisions.

Compensation
No agent recommends, sets, or adjusts pay, bonus, or equity. Comp decisions are human, through your company's existing process.
Termination
No agent flags, recommends, or automates termination. No "PIP risk score." No auto-routed exit packages. Termination stays human end to end.
Promotion
No agent assigns levels, recommends promotions, or auto-fills a calibration committee. The agent can surface signals. The human decides what they mean.
Performance ratings
No agent issues a numeric rating, a performance label, or a stack rank on a human. The Feedback Scorer rates feedback text, not people.
PIPs and corrective action
No agent drafts or files a performance improvement plan. A manager and HR write the PIP. The agent can help assemble prior feedback as source material, with the manager always in the loop.
Conflict resolution and grievance
No agent mediates interpersonal conflict, runs investigations, or generates findings. Those are human processes with explicit confidentiality and legal posture.
Hiring
No agent screens, scores, or ranks candidates inside Branco. We do not ship a hiring product.
Flight-risk prediction
No agent produces an individual flight-risk or attrition prediction score visible to managers.
Individual sentiment scoring
No agent generates a sentiment, mood, or engagement signal attributable to a named person.

This list is the public version of an internal product spec. If we ever consider crossing it, the change ships with a public update to this page first, with a named owner and a comment window.

What each agent sees — and does not see

Branco runs AI inference on Google Vertex AI under DPA terms that prohibit training on Branco's inputs. Our full sub-processor list is published at branco.ai/sub-processors.

Feedback Gatherer
Sees Who is paired with whom, the OKR / skill / competency tag, the prompt text, the response text. Reads only free/busy calendar data via OAuth — never event body, descriptions, attendees, or subject lines. Reads user-and-team mapping (user ID, team ID, language, time zone) needed to deliver a prompt.
Does not see Private DMs, unrelated channels, comp data, performance ratings, role history, manager notes, or any other HRIS field.
Retention Prompt and response stored under your company's standard feedback retention policy. Routing metadata retained 90 days for audit, then dropped.
Access Scoped to the company tenant. No cross-tenant data flow. Every call logged with tenant ID and request ID. Admin on/off is SSO-gated and RBAC-controlled.
Feedback Scorer
Sees The feedback text and the OKR / skill / competency tag. No user identifiers, name fields, or email addresses — identifiers are stripped in the prompt-construction layer.
Does not see Other feedback about the same person, or the identity of the giver or recipient at scoring time.
Retention Feedback text is sent to the inference endpoint to produce the score and is not persisted beyond that call. The score is stored; the text is not retained in inference logs. No training on customer data.
Access The score is visible to the giver as a coaching nudge and to managers in aggregate across a calibration cohort — not as a per-person trend on a single IC.
Action Planner
Sees Feedback the recipient has chosen to share with the agent, their career step, and their company's competency definitions.
Does not see Feedback the recipient has not shared, comp data, calibration discussion, or manager-only notes.
Retention Suggested steps and Small Wins are stored with the plan. Rejected suggestions are dropped within 30 days.
Access The plan is private by default. The human can choose to share with a manager or coach. Sharing is per-plan, never bulk.

Full DPA available on request to procurement@branco.ai.

What changes if you turn agents off

Some customers will turn the agents off. That is supported, intentionally.

Still works

  • Feedback cycle runs on schedule
  • Pairings set manually or by rule
  • Calibration, reviews, action plans
  • Small Wins tracking
  • All your data
  • Full data export via API or CSV/JSON

You lose

  • Feedback quality coaching nudge
  • AI-drafted Action Plan suggestions

There is no AI lock-in inside Branco. The agents are a layer, not a foundation. Agent outputs are stored as annotations on the underlying records — removing them leaves your data intact.

On data portability: all Branco data is exportable via API or scheduled JSON/CSV dump. On termination, a full export is delivered within 30 days at no cost.

What we learned from 2024

In August 2024, an HR-tech vendor announced support for managing AI agents as "digital workers" inside their performance product. The framing was rejected within weeks. The industry has since stayed on the "AI as assistant" side of the line.

We took two lessons from that cycle.

Category language matters. Agents are not employees. They do not have careers, comp, or due process rights. Pretending otherwise breaks the human's relationship to their own performance system. We will not use that framing — in product or in marketing. We hold to the human-as-protagonist frame.

Structural commitments outlast editorial care. A buyer, a DPO, or a reporter who wants to know where Branco stands needs a page they can read in five minutes and forward to a lawyer. That page is this one.

We do not say we are better than anyone in the industry. We say: the industry learned something in 2024, and this page is proof that we learned it too.

The commitments

These are signed and dated. They survive being lifted onto a procurement checklist.

Commitment 1
No AI decisions about humans

Compensation, termination, promotion, ratings, PIPs, conflict resolution, and hiring stay human. Branco will not ship an agent that makes any of these decisions.

Effective 2026-05-19 · Owner: CEO + General Counsel

Commitment 2
Inference-only, no training, named providers

Branco does not train, fine-tune, or RLHF foundation models on customer data. We use third-party inference providers (currently Google Vertex AI) under contracts that prohibit training on Branco-submitted prompts, run in zero-retention mode, and maintain tenant-isolation guarantees. If a provider changes, this page records the new provider before traffic moves.

Effective 2026-05-19 · Owner: CTO

Commitment 3
Agent-off parity

Every Branco feature that matters runs without the agents. Turning them off costs you speed and a quality signal, not the platform.

Effective 2026-05-19 · Owner: Head of Product

Commitment 4
Scoped per-tenant access

Agents operate inside a single company tenant. No cross-tenant data flow. No shared embeddings. No aggregate models built on customer content.

Effective 2026-05-19 · Owner: CTO

Commitment 5
Public changelog

Any change to these commitments ships with a dated update to this page before it ships in product.

Effective 2026-05-19 · Owner: VP Trust / DPO

Commitment 6
Named human owner per decision

Inside the product, every decision the agent does not make is mapped to a named human role (manager, HR, calibration committee). If the role is empty, the workflow blocks.

Effective 2026-05-19 · Owner: Head of Product

If a commitment above ever needs to change, the change appears here first, with a date and a reason.

Frequently asked questions

Is it safe to use Branco?

Yes. Branco's AI agents help your team give better feedback and draft growth plans. They do not make decisions about humans. The six commitments above are the structural reason that is true.

Does Branco let AI decide who gets fired?

No. Termination is a human decision, end to end. No agent flags PIP risk, drafts an exit, or auto-routes corrective action.

Does Branco's AI set compensation?

No. No agent recommends, sets, or adjusts comp, bonus, or equity. Comp decisions go through your existing process.

Does Branco train its models on our feedback data?

No. We do not train foundation models on customer data. We use models scoped to your tenant. Your text stays your text.

Can you turn off Branco's AI agents?

Yes. All three agents can be turned off individually or together from your admin settings. The platform keeps running.

Does Branco rate AI agents?

No. Branco rates feedback signal quality. It does not rate agents on a career ladder, because agents do not have careers.

Who is accountable for these commitments?

Branco, as a company. This page is dated and signed by the CEO. Changes ship here first with a named functional owner.

Stefano Grossi
CEO, Branco
19 May 2026

Changelog

2026-05-19 — First publication. Commitments 1–6 effective.