Some you’ll have considered. Some you’ll have ruled out. One will land harder than the rest. Find that one.
Soma, this is for you. Built around the conversation we had on 25 March, not a deck dressed up with the Unobravo logo.
You were direct about three things. AI is moving faster than the team can absorb if we don’t do this thoughtfully. The rollout has to align with how Paolo and the leadership team are running the business. And the fluency capability needs to live inside Unobravo’s HR career frameworks, not run alongside as a side initiative. You also wanted to identify the internal champions before we deploy widely, and you wanted Ana’s compliance workstream and the EU AI Act readiness sitting next to the programme, not bolted on at the end.
What follows is thirty-six AI agents I’d build with you and your team, each one shaped around a KPI you actually carry as Unobravo’s CPO. Six rows, one per KPI. Six columns, one per mechanism. Each cell is a single agent that does one specific thing on demand and stops. None of them schedules itself or sits in the background. You invoke one, you get one finished output, you decide what happens next.
One of them is marked Start here. That is The Fluency Assessment, the agent we’d build live with you on our next call, end to end, free, mapped to your career framework. It’s the one you asked me for in March. Pick a different one if a different one calls to you louder. The point of this matrix is to make the next conversation concrete before we have it.
See you soon. Mary
One agent. The single one on this matrix you’d notice if it disappeared next month.
That’s the one we’ll build for you. Fully, end to end, at no cost. It runs inside Unobravo, on your stack, with your career framework. Your team learns how it was made, and it stays with you whether or not we work together again.
We’ve marked one cell Start here: The Fluency Assessment. It’s the agent you specifically asked me for on 25 March, the one that plugs into your HR career frameworks and makes "what level is this person at" answerable in evidence rather than opinion. It’s a hint, not a verdict. Pick whichever you’d notice the most.
The other thirty-five are open conversations. Some are obvious next builds for us to do alongside you. Some are perfect first agents for your team to build themselves, and the AI fluency programme we discussed is designed for exactly that. Either route ends with the capability sitting inside Unobravo, not with us.
The agents we build run inside Unobravo, on Google Workspace where you stand today, with your data. We don’t store, train on, or share that data. Therapist conversations, patient records and clinical data sit behind your existing access controls and never enter our environment. When the engagement ends, the capability stays inside your business; no copy travels with us.
Are those six the KPIs you actually carry? If a row feels off, tell us. We drew them from your remit as CPO, your public moves, and what you and Ana said on 25 March. You know the real list.
Six mechanisms, defined below the matrix. Each is a way one agent, run once, produces one specific deliverable. None of them is autonomous. All of them stop after one output.
The cell where the row matters most and the mechanism feels most natural. Read its detailed spec underneath. Tell us why that one and we’ll build it live on our next call.
Take an input that takes hours of human work and return the finished artefact in minutes. One run, one output, same fidelity.
Apply one senior’s judgement, voice or expertise to one specific input. Each invocation produces one tailored output in their style.
Take one specific case or situation and return one forecast of how it will resolve, with reasoning. Forward-looking by design.
Take one recipient’s context and return one output shaped to that recipient. Each invocation is one-to-one.
Examine one specific artefact, a draft, a record, a thread, a file, and return one findings report flagging issues, gaps or risks.
Take one decision context and return one ranked recommendation pack with reasoning. Not autonomy. One-shot advice for a human approver.
Each spec includes the mechanism it uses, the KPI movement we’d expect, our reasoning trace from your business and our conversation, and the smallest first build, the version we’d ship to prove the rest.
You said fluency has to live inside how the business runs, not as a programme bolted on top. That means agents that translate "AI literacy" into role-specific, role-recognisable work. Not awareness sessions. Useful tools per role, in the rhythm of the job.
From one job title at Unobravo, returns a one-page brief: what AI fluency looks like for this role, the top three workflows AI changes, the three prompts that pay off first, what good looks like at level 2 versus level 3.
Given a colleague’s question about a prompt, a workflow or a tool, returns one answer in the voice of one of your internal AI champions, drawing on what they’ve already taught and how they’ve already explained it.
Given a department’s recent training activity, prompt-library usage and self-reported fluency, returns a forecast of which teams are ready to step up a fluency level and which need another pass.
From one Unobraver’s role, current level and recent work, returns a tailored 30-day pathway with the specific prompts, gems and workflows that fit their day-to-day, not a generic curriculum.
Examines your central AI knowledge hub, the Notion prompt library, gems, shared docs, and returns one report flagging stale prompts, duplicate gems, missing patterns by function, and the five prompts that earn their keep.
From the full Unobraver list, role data and engagement signals, returns one ranked pack: which thirty people to put through the next cohort and why each one, with the leadership intent it serves.
A regulated mental-health platform with cross-border operations doesn’t get to treat the AI Act as a side project. You asked Ana to run the readiness work alongside the fluency programme. These six agents make the compliance side as concrete as the training side.
Take one Article of the EU AI Act and return a one-page Unobravo-specific brief: who in the business this hits, what we need to do, what the evidence looks like, what the deadline is in our terms.
Given an internal question about an AI use case ("can we use Gemini to summarise patient feedback?"), returns one answer in your compliance team’s voice, drawing on Unobravo’s actual policies and the AI Act’s positions.
Given one proposed internal AI use case (free text), returns a risk forecast under the AI Act with reasoning, the likely classification (minimal, limited, high), and the gating evidence that would change the classification.
From one employee’s role, returns one tailored AI Act training note: what they need to know, what they don’t, by the systems they actually touch. Therapists get a different note from marketing leads.
Given the current training record from your HR system, returns one report flagging who is owed AI Act training, by when, why, and the deadline relative to your obligations under Article 4.
From the catalogue of internal AI uses, returns one ranked pack: which uses to formalise into the risk register, which to monitor lightly, which to restrict, which to retire. Reasoned per item.
Five thousand therapists on the platform today and growing. International expansion turns each new country into a multilingual hiring problem. These agents lift the velocity without flattening the clinical care your brand depends on.
From one candidate’s CV, returns a one-page brief tuned to Unobravo’s clinical standards: training fit, language coverage, supervision needs, modality match, red flags, opening questions for the screening call.
Given one inbound application or candidate query, returns one response in your senior therapist recruiter’s voice, with the warmth your brand requires and the clinical specificity the candidate needs.
Given one open role’s language, seniority and location, returns a forecast of how long it will take to fill, what the funnel will look like at each stage, and where it’s likeliest to leak.
From one named target therapist, returns one tailored outreach in their language and clinical specialism, referencing Unobravo’s care model and the actual reason they’d move (not the generic one).
Given the current therapist pipeline, returns one report on language coverage, seniority spread, clinical modality mix and gaps by market. Names the three things to fix before the next expansion review.
From the international expansion plan and current platform demand, returns one ranked pack: which three languages and two seniority bands to source against this quarter, with reasoning.
A scaling platform hires faster than its onboarding can absorb. Every week of slow ramp is a week of paid silence. These agents make the first thirty days personal, observable, and recoverable.
Given a new joiner’s role, language and start date, returns a one-page Day-One pack: who to meet, what to read, the three workflows to learn first, the one prompt that will save them an hour this week.
Given a new joiner’s question, returns one answer in the voice of their assigned buddy, drawing on the buddy’s own internal docs and shared answers, so the joiner gets help even when the buddy is in session with a patient.
Given a new joiner’s first thirty days of activity (training, platform usage, manager touchpoints), returns a forecast of whether they’re tracking on, ahead of, or behind ramp, with reasons and the specific intervention that would shift it.
From one joiner’s role, language, prior experience and stated development goals, returns one tailored onboarding pathway, week by week, with named milestones and resources in their language.
Given a new joiner’s first thirty days, returns one report on which milestones they’ve hit, missed, or skipped, with the specific reason and the manager moment that should have caught it.
From the current onboarding cohort, returns one ranked pack of who needs a check-in this week and on what specifically, with the manager moment most likely to land.
The thing you said you wanted most on 25 March: AI fluency that plugs into how Unobravo already promotes, develops and rewards people. Not a parallel track. The career framework, made AI-aware. This row, and especially The Fluency Assessment, is where the programme earns the right to live.
From one employee’s last review, recent work product and current career-framework level, returns a one-page brief for their manager: where they are, where to push next, two questions to open the conversation with.
Given a development question from a direct report, returns one answer in your best coaching manager’s voice, mapped to your career framework and grounded in the specific evidence from the report’s recent work.
Given six months of one employee’s work signals, peer feedback and career-framework benchmarks, returns a forecast on readiness for the next career step, with the specific gaps and what would close them.
From one employee and one named skill (technical, clinical, leadership, AI), returns one tailored development pathway with internal mentors, named resources, milestones and the workflow it changes.
Given one employee’s recent AI work product (prompts, gems, workflows, outputs), returns one assessment against your AI fluency rubric: where they sit on the level 1 to 4 ladder, with named evidence, the gap to the next level, and the two practices that would close it.
From a team’s last quarter of work signals, returns one ranked pack: who to promote, who to stretch, who to coach harder, who to move, with reasoning the leadership team can defend.
A B Corp mental-health platform sells care. Internal care is the same product. Paolo named the B Corp impact narrative as a frame for this whole investment. These agents make the people story tangible, measurable and tellable.
From last month’s engagement signals (survey, Slack, manager touchpoints, recognition), returns a one-page brief on what’s working, what’s slipping, and the three things to act on this week.
Given one internal moment, a launch, an incident, a milestone, returns one written communication in Danila’s voice that lands with the team and stays true to what she’d actually say.
Given one employee’s recent signals, returns a forecast of retention risk with the specific reasons it’s rising and the intervention most likely to shift it before notice is in.
Given one employee, returns one personal anniversary note that names their actual work, their actual impact and their actual relationships, not a generic thank-you with their name swapped in.
Given the last quarter’s people moments, returns one report on which stories belong in the B Corp impact report, what evidence they need, and how to tell them in a way the certification body will accept.
From the at-risk list, returns one ranked pack of three actions: who from leadership reaches out, with what message, by when, and what success would look like.