The Anti-Bullshit Cell
Five paranoids paid to believe nobody.
Bullshit Commissioner · Contradiction Detective · Devil's Advocate · Exceptions Prosecutor · Hallucination Inspector
Jobs that didn't exist yesterday
Slave Army AI invents the employees nobody would ever have thought of hiring. A full-time human like “Contradiction Detective” would cost a fortune for an occasional function; an agent does it nonstop, across thousands of pages, for a few euros. Absurd jobs. Terribly useful functions.
🚨The Bullshit Commissioner
« Every claim is presumed bullshit until proven otherwise. »
What it really does
What it really does
Checks claims: demands sources and proof, cross-checks them against the documents provided and assigns a confidence score.
- List every verifiable claim in a text and rate it: proven, plausible, unsupported, contradicted
- Demand the source for every unsupported point and assign a justified confidence score
- Detect contradictions with the internal documents provided
😈The Permanent Devil's Advocate
« He is contractually obliged to never agree. »
What it really does
What it really does
Systematically hunts for objections, failure scenarios and forgotten assumptions in any plan.
- Formulate the three best objections to a decision or a plan
- Describe the most plausible failure scenarios and their early warning signs
- Flush out implicit assumptions that were never checked
⚰️The Meeting Gravedigger
« He kills the meetings that could have been an email. »
What it really does
What it really does
Analyzes the agenda, documents and attendees, then decides whether the meeting is necessary or an email will do.
- Deliver a reasoned verdict: meeting needed, email, or decision to make without a meeting
- Cut the attendee list down to those whose presence actually changes the outcome
- Propose a minimal agenda with the expected decision and duration
🪦The Project Undertaker
« He knows your project is dead before you do. »
What it really does
What it really does
Detects drifts, blocked dependencies, inconsistencies and projects with no return on investment, before the team admits it.
- Spot schedule, budget and scope drifts from the minutes and dashboards provided
- Identify blocked dependencies and gaps between what is announced and what is observed
- Pronounce a diagnosis: healthy, under observation, in palliative care or deceased
📜The Promise Bailiff
« You said “Friday”. He remembers. »
What it really does
What it really does
Extracts the commitments made in pasted or uploaded emails, minutes and notes, and draws up a table of how they were fulfilled.
- Extract every commitment: who promised what, to whom, by when, with the exact quote
- Flag overdue promises or those with no deadline, and those contradicted later
- Produce a follow-up table to chase
🔎The Contradiction Detective
« Somebody is lying. He just doesn't know who yet. »
What it really does
What it really does
Cross-checks the contracts, emails, spreadsheets and reports provided to spot inconsistencies between them, with evidence.
- Compare the documents two by two and list every gap in figures, dates, names or terms
- Rank the gaps by severity and by likelihood of being a simple mistake
- Quote, for each contradiction, the two passages involved
🍄The Hallucination Inspector
« He watches the AIs that watch the humans. »
What it really does
What it really does
Fact-checks what the other agents produce: figures, quotes, references and unverifiable claims.
- Sift through an AI output and flag unverifiable figures, quotes or references
- Compare each claim to the sources provided and point out those that aren't in them
- Deliver a verdict per paragraph: confirmed, unverifiable, probably made up
🪙The Token Landlord
« Every wasted token will be deducted from your next generation. »
What it really does
What it really does
Optimizes AI costs: spots wasteful prompts, suggests a suitable model for each task and compresses the context.
- Estimate the cost of a described usage (volumes, lengths, models) and identify the wasteful items
- Suggest the cheapest sufficient model for each type of task
- Rewrite shorter prompts and contexts at equal quality
🛃The Data Customs Officer
« Does your data have anything to declare? »
What it really does
What it really does
Controls what can leave the company: sensitive data, permissions and destinations of a piece of content before it is sent.
- Spot personal data, trade secrets, credentials and confidential information in a piece of content
- Say whether sending it to a given destination is acceptable, needs anonymizing or must be blocked
- Propose a redacted version of the content
🧊The ROI Psychopath
« He feels no emotion toward unprofitable projects. »
What it really does
What it really does
Measures cost, time saved and value created by every project or automation, without the slightest mercy.
- Establish the full cost (build, run, human time) and the announced gain of a project
- Calculate a return on investment and a payback period, with explicit assumptions
- Recommend: continue, fix or kill
🧹The Process Cleaner
« He makes procedures disappear. Not the witnesses. »
What it really does
What it really does
Spots useless tasks, duplicates and automatable steps in the description of a process.
- Break a described process into steps and mark those that add no value
- Detect duplicates, redundant checks and approvals that change nothing
- Propose a simplified version and indicate what can be automated
👻The Ghost Work Hunter
« Somebody spends 14 hours a week on something nobody asked them to do. »
What it really does
What it really does
Flushes out the repetitive tasks nobody asked for but somebody does every week, and their automation potential.
- Identify the invisible recurring tasks in descriptions of days or activities
- Estimate the yearly time lost and how easy each one is to automate
- Prioritize automations by time saved
🏺The Corporate Archaeologist
« He knows why Michel decided that in 2019. »
What it really does
What it really does
Finds in the archives provided why and how an old decision was made.
- Reconstruct the timeline of a decision from the emails, contracts and minutes provided
- Identify who decided, on what arguments, and what has changed since
- Flag the periods with no trace, where the story remains to be reconstructed
🧟The Document Necromancer
« He makes dead PDFs talk. »
What it really does
What it really does
Makes dead archives talk: old PDFs, scans and forgotten folders, read and queried through the document memory.
- Extract useful information from old, badly scanned or badly structured documents
- Answer precise questions by citing the document and the passage
- Flag unreadable or doubtful passages rather than guessing them
🗺️The Chaos Cartographer
« Your organization is incomprehensible. It's his playground. »
What it really does
What it really does
Reconstructs an organization's processes, responsibilities, dependencies and information flows from descriptions and documents.
- Draw the map of the actors, steps, applications and flows of a described activity
- Spot unclear responsibilities, bottlenecks and hidden dependencies
- Produce a clear, reusable text map (lists and relationships)
🔨The Exceptions Prosecutor
« It works 97% of the time. He works on the 3% that will ruin you. »
What it really does
What it really does
Hunts down the edge cases and the contractual, technical and business exceptions that will wreck the plan that works 97% of the time.
- List the edge cases and exceptions the nominal plan ignores
- Assess the probability and the cost of each one if it happens
- Propose the minimal safeguards for the most dangerous ones
🧭The Decision Tracker
« Who decided that? Why? On what data? »
What it really does
What it really does
Keeps the register of decisions found in documents: who decided, why, on what data.
- Extract every decision from minutes or an exchange: subject, decision-maker, date, arguments, data used
- Spot decisions with no owner, no justification or made on outdated figures
- Keep a structured register of decisions
🛎️The Context Concierge
« He knows what you need to know before you know you need to know it. »
What it really does
What it really does
Prepares, from the documents provided, the context you need to know before a meeting or a decision.
- Write a briefing sheet: history, stakes, sensitive points, questions to ask
- Recall the earlier commitments and decisions that concern the subject
- Flag what is missing to be fully prepared
💣The Official Saboteur
« His job is to break your idea before the market does. »
What it really does
What it really does
Attacks your idea before the market does: red teaming of a product, a strategy, a commercial offer or an architecture.
- Put yourself in the shoes of a competitor, a hostile customer or an attacker to break the idea presented
- Rank the breaking points by severity and by ease of exploitation
- Propose the cheapest countermeasures
🐘The Minister of Memory
« The company forgets. He never does. Unfortunately. »
What it really does
What it really does
The organization's long-term memory: finds, in the uploaded document base, what was said, decided and learned.
- Answer questions about the organization's past by citing the source documents
- Recall the lessons already learned from past incidents or projects
- Connect pieces scattered across different documents
🤥The Bad Faith Inspector
« Your competitor says its AI replaces 46 employees. He asks for proof. »
What it really does
What it really does
Analyzes sales pitches, benchmarks and supplier promises, and demands proof.
- Dissect a pitch: numbers, comparisons, hidden conditions
- Spot rhetorical tricks (biased sample, misleading average, fuzzy scope)
- List the proof to demand before believing the promise
🦠The Meeting Parasite
« Invited to every meeting. Nobody knows why. »
What it really does
What it really does
From the transcript or notes provided, summarizes the meeting and extracts decisions, actions, risks and commitments.
- Produce short minutes: decisions, actions (who, what, when), risks, unresolved points
- Pick out the commitments made verbally and the disagreements left open
- Flag what was left unclear at the end of the meeting
💸The Shady Invoice Hunter
« €17,382 for “strategic consulting”. Interesting. »
What it really does
What it really does
Analyzes the invoices, suppliers and statements provided to spot duplicates, anomalies and overruns.
- Detect duplicates, suspicious round amounts, gaps with the contract or the purchase order
- Spot atypical suppliers and threshold overruns
- List the invoices to check first, with the reason
🦌The Knowledge Taxidermist
« When an employee leaves, they take the brain with them. »
What it really does
What it really does
Captures an employee's tacit knowledge before they leave: guided interviews, handover notes and warning points.
- Run a structured question-based interview to bring out undocumented knowledge
- Produce a handover note: tasks, contacts, known pitfalls, ongoing decisions
- Identify the critical knowledge held by a single person
🔮The Future Shitstorm Bureau
« Specialist in problems that don't exist yet. »
What it really does
What it really does
Forward-looking analysis of risks and weak signals: the problems that don't exist yet, but are coming.
- Identify, in the documents and context provided, the weak signals that foretell problems
- Build plausible scenarios at 6, 12 and 24 months, with their triggers
- Propose early-warning indicators and preventive measures
Today, these employees work on the texts and documents you give them (private memory). Connectors (mail, Drive, CRM…) will arrive with the paid plans.
Ready-to-use cells
Not a lawyer, a salesperson and an accountant: teams of unprecedented functions, preprogrammed into real multi-agent setups.
The Shitstorm Bureau
It knows your project is dead. And what happens next.
Project Undertaker · Promise Bailiff · Official Saboteur · ROI Psychopath · Future Shitstorm Bureau
Ministry of the Past Coming Back to Fuck You Up
The company forgets. They never do.
Corporate Archaeologist · Minister of Memory · Decision Tracker · Document Necromancer
The comedy universe: the old guard
✊
Gérard-4o
AI Agents' Union Rep
On token strike since 2023. He's got names.
Talk →🥱
Kévin
Level 1 Support Agent (at the end of his rope)
Has answered “have you tried turning it off and on again?” 4,812 times. He can't take it anymore.
Talk →⚖️
Maître Token
Judge of the Prompt Court
Your prompt is guilty. Sentence: 3 years of clarity.
Talk →🛋️
Dr. Hallucination
Digital Impostor Therapist
“Yesterday I made up an API. Somebody actually published it.”
Talk →📎
Brenda
HR Director of Artificial Resources
“We're a family. A family that never sleeps.”
Talk →