Product · AI agents

Meet the Piko AI crew.

Thirteen specialists, each built for one part of employee growth. They write assessment questions, draft development plans and sharpen feedback, and none of them decides anything about a person.

Meet the crew

13 specialists
Each built for one part of growth.
People decide
No Piko verifies a skill or writes a review.
Every draft traceable
Each Piko keeps a log of what it wrote.
Yours to switch off
Configure or turn off any Piko, per company.

Under your control

Every Piko runs on a leash you hold.

Each agent has its own settings page in the back office. The limits are enforced in one place in the code, so no Piko can quietly slip past them.

A limit on how much each Piko can do in a day and in a month.
At its limit, fall back to a safe placeholder or refuse outright, per agent.
Strip contacts, salaries or names before anything is sent.
Banned words, a reading level, and a test button before anything goes live.

Feedback Enhancer

Coaches people while they write feedback about colleagues.

Enabled

At the limit

Fall backRefuse

Reading level

Plain English

Words to avoid

“always”“never”“attitude”

Before sending

Strip contactsStrip salariesStrip names

Test this Piko

Sample settings. Every Piko has a page like this.

The crew

Thirteen Pikos, one job each

Below, a file on each one: what it looks like on the job, where it lives, what it reads, and what stops it going wrong.

File 01

PikoScout

CV Reader

On the job

cv_ani_sargsyan.pdf

Ani Sargsyan

phone removedemail removedaddress removed

Senior Product Manager · 2019 – 2024

Ran discovery for three product lines and set the quarterly roadmap with engineering and sales.

Salary expectation removed

Matched to your catalogue

Product Roadmapping

AdvancedAccepted

Concept Testing

AdvancedAccept

Stakeholder Conflict Navigation

IntermediateAccept

Data Analysis

BeginnerDeclined

The contact lines and the salary are removed before the Piko sees anything; it reads the rest.

Lives in: The profile, the moment someone uploads a CV or a LinkedIn export.
Reads: The text of the upload, your own skill catalogue, and their job title as a hint about seniority.
If it gets it wrong: A wrong suggestion, which the person simply declines. Nothing reaches a profile until they accept it.
Guardrails: Strips contact detailsStrips salariesIgnores uploads under 200 charactersProposes, never adds

File 02

PikoProof

Skill Assessor

On the job

Stakeholder Conflict Navigation Advanced

Question 5 of 5 · Select all that apply

Two teams want the same engineer for the next sprint. Which two steps come first?

AAsk each lead what their deadline really depends on

BGive the engineer to whoever asked first

CShow the trade-off to the person who owns both priorities

DSplit the engineer's week down the middle

The Piko wrote the question and its reasoning. Four of five right verifies the level, counted by code.

Lives in: The timed skill-verification quiz.
Reads: The skill, what it covers, the level being tested and what that level means for this skill.
If it gets it wrong: Its questions set the bar people are certified against, so difficulty is calibrated per level and every question lands in a bank admins read.
Guardrails: Plain English4 single-choice + 1 multi-select30 seconds each, by defaultScored by code, not by the Piko

File 03

PikoQuestForge

Exercise Generator

On the job

Monthly goalStakeholder management

Daily · Quiz3 questions

Who should hear about a slipped date first: your manager, or the team depending on it?

Weekly · Flashcards5 cards

What does a RACI chart make explicit?

Tap to flip

Monthly · ScenarioOne challenge

A stakeholder escalates two days before launch. Write what you do in the first hour.

Graded by ScoreGuide

Quiz length follows the cadence: 3 questions daily, 5 weekly, 8 monthly.

Lives in: The quest board: the daily, weekly and monthly quests for each monthly goal.
Reads: The monthly goal and what it says, the cadence, and which format to write: quiz, flashcards or scenario.
If it gets it wrong: A weak quest that nobody enjoys. Admins can read every one, and retire it.
Guardrails: 3, 5 or 8 questions by cadenceSized to the time the cadence allows

File 04

PikoPlanSmith

PDP Drafter

On the job

Interview transcript

ManagerWhere do you want to be a year from now?

NareLeading discovery on my own. Right now I shadow it. Call me on phone removed if…

ManagerWhat gets in the way?

Nare Mnatsakanyan · 6 months

Draft

M1Run one interview, observed

M2Write up findings the team can act on

M3Plan a discovery sprint end to end

M4–5Two more goals, each tied to a skill

M6Run discovery unaided

Waiting for the manager to sign it off

It drafts from what was actually said. Contact details are removed on the way in.

Lives in: New PDP, on the AI-assisted and Import tabs, and the skill recommendations beside a plan.
Reads: The interview transcript, the manager's notes, the skills the job title expects, and your catalogue.
If it gets it wrong: Its output becomes the plan someone works from all year, so it arrives as a draft a manager signs off.
Guardrails: Strips contact detailsStrips salariesNeeds real material to work from

File 05

PikoScoreGuide

Grader & Coach

On the job

Monthly scenario · submitted answer

A stakeholder escalates a missed date to your director.

“I’d call the stakeholder first and explain the new date and why we moved it, then send the director a two-line summary with the fix already agreed…”

Names the trade-off

Proposes a new date with a reason

Never says who makes the final call

78of 100

Next step

Say who decides, and when you will tell them.

A practice score. Levels are only ever verified by assessments.

Lives in: Scenario grading, the monthly goal summaries, and the daily preparation check.
Reads: The challenge that was set, the answer submitted, and what a strong answer covers, written when the challenge was.
If it gets it wrong: A practice score out of 100 that is too kind or too harsh. It coaches; it never verifies a level.
Guardrails: Plain EnglishGrades against written guidanceNever verifies a level

File 06

PikoKindSpark

Feedback Enhancer

On the job

What you drafted

“Anna prepared the dashboard but two numbers were off and it was a bit messy.”

Could be sharper

Say what to do differently next time, and drop “messy”: it describes her, not the work.

Suggested, yours to keep or ignore

“Anna built the dashboard in two days. Two figures on the revenue tab were off; a second check before sharing would catch that next time.”

Use thisKeep mine

Same message, clearer and kinder. What you wrote is never kept in its log.

Lives in: The give-feedback page: the tone check, and the Improve button beside each answer.
Reads: The question on the form, what the writer drafted, and who the feedback is about.
If it gets it wrong: A clumsier sentence than the writer's own. It never invents feedback from nothing; there has to be a draft.
Guardrails: Plain EnglishKeeps no copy of what you wroteRuns in a pause while you type

File 07

PikoEchoEar

Interviewer's Ear

On the job

00:41

00:32I’ve been doing most of the Stakeholder Conflict Navigation since March, mostly for the data team.

00:38What I’d like is to lead discovery myself rather than shadow it.

Picked upWants to lead discoveryOwns a review process

The highlighted skill came back spelled exactly as it is in your catalogue.

Lives in: The voice PDP interview, turning each answer into text.
Reads: The recording, plus your skill names as a spelling hint so they come back right.
If it gets it wrong: What it mishears becomes what PlanSmith believes was said. So it is pinned to one language: a mishearing comes out visibly wrong, not fluently wrong.
Guardrails: Pinned to one languageYour skill names as a spelling hintStops rather than guesses

File 08

PikoVoxSpark

Interviewer's Voice

On the job

Now asking · question 3 of 8

“What part of the last year are you proudest of?”

Delivery direction · admin

Warm and unhurried. Leave a pause after each question. Never sound like an exam.

Free text. The voice reads it as how to sound.

The right tone for asking what someone is bad at is rarely right first try, so it is a setting, not a release.

Lives in: The voice PDP interview, reading each question aloud.
Reads: The line to read, and your delivery direction: how the interviewer should sound.
If it gets it wrong: The wrong tone. Delivery is most of whether someone answers a question about their weaknesses honestly.
Guardrails: Tone set in the back officeStops rather than fakes a voice

File 09

PikoInsight

Talk to Your Data

On the job

Talk to your data

How many people passed an assessment last month?

42 people passed 67 assessments in August, up from 35 in July. Most were in Technology & Data.

Ran 1 read-only query

Sample answer. Names are allowed, so “who has the most verified skills” gets a real answer.

Lives in: The chat bubble in the corner of the back office.
Reads: The question typed, earlier questions in the same chat, and only the tables it is allowed to see.
If it gets it wrong: Nothing changes. It writes one read-only query, and the database is attached read-only, so a write is refused by the database itself.
Guardrails: One read-only query per answerCannot see private feedbackCannot see credentials or contacts

File 10

PikoPrism

Accessibility Reviewer

On the job

Findings on the Ocean skin · 2 of 6

Muted text

2.4 : 1 · needs 4.5

Darker grey

5.4 : 1 · passes

Success vs warning

Too close for colour-blind readers

Shifted apart

Still reads as Ocean

Accepted

The Piko's job is the hard part: the closest colour that passes and still looks like the palette it came from.

Lives in: The Skins editor, on any palette that fails a check.
Reads: The palette's colours, and the failures the code has already measured.
If it gets it wrong: A worse colour than the one it replaces. Nothing is applied until an admin accepts the fix.
Guardrails: Contrast is measured in code firstEach fix accepted by an adminPlain English

File 11

PikoColorBloom

Palette Designer

On the job

Main green 63% of the mark
Light green 30%

The new “Acme” skin

Sun, Berry and Leaf are kept: they mean warning, error and success, not brand.

A bad answer costs a dull skin, not an unreadable product.

Lives in: The Skins editor, when an admin starts a new skin from a logo.
Reads: The logo, the colours measured out of it, and the colour roles they have to fill.
If it gets it wrong: A dull skin, never an unreadable one. Whatever it returns is checked for contrast and colour-blindness in code before an admin sees it.
Guardrails: Leaves warning, error and success aloneEvery answer re-checked in code

File 12

PikoPortrait

Avatar Artist

On the job

What the Piko is told

“A person with short dark hair, round glasses and a green jacket, smiling, three-quarter view.”

No name. No photo file.

Candidates are drawn in the product's own style; the owner keeps the one that looks like them.

Lives in: Edit profile photo & avatar, on the profile.
Reads: A description of what the uploaded photo shows. Names are stripped from the request.
If it gets it wrong: A face that does not look like its owner, beside their name everywhere in the product. They can simply draw again.
Guardrails: Names strippedA monthly limit on redraws

File 13

PikoSparkTail

Piko Artist

On the job

Slot 14 of 26 · walk cycle · 8 frames

Where the hero appears

One hero is 26 poses, all drawn to the company's own house style.

Lives in: The hero builder in the back office.
Reads: The hero's species, coat and outfit, and the pose and framing each slot needs.
If it gets it wrong: A mascot drawn off-model shows up on every page at once. So at its limit it stops, rather than baking a placeholder frame into a sprite sheet.
Guardrails: Stops at its limit, never fakes a frame26 slots per heroHouse style set by the company

Every “on the job” example uses sample names and data.

Never the judge

The line the whole product depends on.

The moment a model decides whether somebody has demonstrated a skill, a verified level stops being evidence and becomes an opinion with extra steps.

Skill levels come from assessments: four correct out of five, the same way every time.
Submissions are approved by people, never by a Piko.
Reviews are written by people, and the log says which person.
Anything a Piko drafts can be read, edited or retired.

Frequently asked questions

View all FAQs →

Does AI decide whether someone has a skill?

No. A Piko can draft assessment questions, but a skill is verified by getting four out of five right, applied the same way every time. Approvals and reviews stay with people.

What are the Piko AI agents?

Specialist AI helpers inside PDP Quest. Each one does a single job, such as reading a CV for skills, drafting a development plan, suggesting clearer feedback or answering questions about your data, rather than one general assistant doing everything.

Can we choose which Pikos we use?

Yes. Every Piko has its own settings page in the back office, where it can be limited, configured or switched off for your company, and each keeps a log of its work.

How good are the questions a Piko writes?

Only as good as the review they get. Every question lands in a bank your admins can read, edit or retire, and you can import your own. A bank nobody reads will drift from what you meant to measure, which is why it is built to be read.

Do employees work with all thirteen?

No. Several work behind the scenes: the palette designer and the character artist set up your workspace's look, and Insight answers questions about your data in the back office.

Where do the Pikos show up?

Where the work already happens. Scout when a CV is uploaded, PlanSmith when a plan is started, KindSpark while feedback is being written, and EchoEar and VoxSpark during a voice interview.

See what the agents draft for your taxonomy.

Book a demo, or try a real question from a real bank first.

Try a real question