Source: Concept.md

Overview

Deepwork is a platform that facilitates real-time human coordination at scale using human agents guided by a hybrid architecture composed of centralized coordinating agent driven by AI (to achieve lowest possible latency while preserving highest possible consistency in achieving optimal strategy) and distributed sensory-motor agents (human, for maximal responsiveness to novel situations, low-latency sensory-motor infrastructure, versatility of applications) to feed in real-time observations about the dynamic environment in which the coordinated outcome is engineered.

Concept

• Deepwork is a company that allows customers to engineer large-scale coordinated outcome in the world. For example, a business man in Chennai decides to ensure that the whole of Marina beach is fully cleaned in the next 6 hours. They come to the Deepwork website and describe the task and provide the clarifications, needed to generate a complete outcome specification along with constraints. Since we're a law-abiding company, by default we will add all legal constraints as part of the base constraints of the specification. Using the outcome spec as input, we will generate a comprehensive dynamic action plan that changes real-time in response to the environment. An action plan is composed of jobs that are generated dynamically using AI.
• Deepwork hires a large number of people at a base minimum hourly pay. They get paid at least that much every hour that they are actively working on jobs. They can choose what job to work on. The company ensures to certify everyone so that jobs only become visible to eligible Deepworkers. Deepworkers can sign up to get certified for various skills. The more skills they certify for or rarer the skill they certify for is, the more niche jobs they can begin to sign up for.
• The key USP of Deepwork is that we will get the job done in the fastest possible time no matter what the job is. This promise depends on us having enough Deepworkers qualified to do the jobs being posted. So we may have to have a coldstart period since this is somewhat of a marketplace. What we're essentially selling is a generalized architecture for optimally contracting human labor. This reduces the cost for the one posting the job and also maximizes the freedom, transparency and consistency for the one looking for a job, since all jobs posted on Deepwork are fully transparent in terms of the terms and individual Deepworkers can apply filters to only list jobs that they are capable of taking that adhere to their terms.
• To faciliate this operation we will allow anyone to create an account on Deepwork and validate their unique identity and then ask them to book a paid training appointment where they will be trained and certified to operate as a Deepworker. They need to prove to us that they are capable of operating on the Deepwork platform. Once they do, they will start seeing all the jobs that they are eligible for, which will most likely be jobs that do not need any skill. And these are going to be going at the lowest possible hourly rate. As each Deepworker certifies for more skills and build their reputation, they get awarded jobs that pay at a higher rate. So it's a career progression of sorts.
• Once the outcome spec is created, our proprietary algorithm dynamically generates jobs with their own constraints (time, quality, etc). The job's instructions change dynamically according to progress from all the other deepworkers doing the job. There are two modes of coordination. One is through protocol-driven manual coordination and there's assisted coordionation. The protocol-driven manual coordination depends on the Deepworker reporting changes dynamically to the coordinating agent. The coordinating agent is just an AI agent that takes in the inputs provided by the Deepworker, investigates it and analyses it and creates reports using the Coredoc protocol to the mental model associated with the job. The job execution plan changes dynamically based on changes to the action plan mental model. The assisted coordination involves a Deepworker helmet that uses AR to make the observational and inferential claims and passes commands to the Deepworker to guide the sensory-motor outcome in the Deepworker.
• We will build and allow franchising of standardized certification labs that Deepworkers can arrive at to get certified on various skills. It's essentially like a highly optimized educational center that trains individuals for various skills. We can even use it as a WRST (Waking Rest Stillness Test) assessment center.
• The skills also have levels and rankings which allow the Deepworker to progress in terms of their competence. We could have whole teams of human workers who follow standard protocols that allows them to transfer across the board.
• Deepwork profits from commissions on the hourly rates, job posting and coordination fees.
• As for the tech, our proprietary technology depends on dynamic construction of mental models that facilitates large scale realtime coordination via distributed jobs and deepworkers.

Core Tech

• The platform itself - just a whitelabeled Clikkin app that allows prospective deepworkers to register and guides them towards identity verification, certifications and jobs and also to commission outcomes.

• The certification centers will have their own admin dashboards in the Deepwork app. Let's say Dave wants to take on carpentry jobs and he signs up to enroll for jobs. The app instructs him to visit the nearest certification center via an appointment and guides him via the map to go there. It also instructs him to dress appropriately, take the necessary documents and guides him with behavioral cues, like "be polite, friendly to other" etc. Once he arrives there, he will be asked to meet "Reeta, the Orienter" wherever her location is listed. Once Dave reaches Reeta, Reeta will administer a feedback survey to Dave and note down the responses. Reeta then asks Dave to follow Mike who arrives to receive Dave and then introduce himself as his observer and guide for the certification session and proceeds to explain how the certification session will be conducted and what Dave's objectives are. In the backend the coordinating agent analyzes the background of the candidate and their stated interests in the kind of jobs they wish to take, creates a customized action plan to take them through the necessary custom certifications needed to be qualified for the jobs. The coordinating agent assigns the necessary tasks to Mike so that he may guide and observe Dave as Dave is asked to complete custom tasks that are part of the certification. This is how the dialogue proceeds

  • Coordinating agent to Mike's device - "<outcome spec with instructions to perform domain-specific tasks>"
  • Coordinating agent to Mike's earpiece - "Instruct Dave to check his device for instructions on what he must do"
  • Mike to Dave - "Please have a look at your device for instructions"
  • Dave to Mike - "Sure, let me check" and proceeds to read the instructions and at the end of the task instructions, it asks Dave to let Mike know that he's read the instructions and thank Mike."
  • Mike to Dave - "I've read the instructions, thanks!"
  • Coordinating agent to Mike's earpiece - "Use your smartphone to record Dave as he performs the task"
  • Coordinating agent - watches Dave perform the task live and uses AI and creates a preliminary analysis and also distributes the video as tasks to other Deepworkers who are qaualified to act as professional evaluators of the tasks performed in that domain. The coordinating agent processes all the scores and executes a standard grading protocol that grades Dave's performance.
  • [If grade was dissatisfactory] Coordinating agent to Mike - "Thank Dave for the completion of task and let him know that he may leave and he'll be notified on the device when jobs will be available for him"
  • [If grade was satisfactory] Coordinating agent to Mike - "Proceed to stall C and retrieve a Deepwork welcome kit and offer it to Dave and ask him to go to the orientation hall and then let him know that he'll receive further instructions on his device once he gets there and thank him for his participation with a smile."
  • Mike to Dave - "Follow me" and he walks to stall C and offers the welcome kit to Dave. "Take this to the orientation hall and open it and you'll receive further instructions on your device. Thank you." The welcome kit contains a wrist device (imagine something the size of a smartphone) with two tactile buttons (one black and one red) and screen to show instructions/text/images/videos etc and an earpiece to both record the audio of the wearer's surroundings and to pass auditory instructions to the wearer.
  • Dave - Walks to the orientation hall and opens the welcome kit. It has a note to scan the QR code with his device. Once the welcome kit is activated on his device, it asks him to wear the wrist device and earpiece to receive further instructions.
  • Coordinating agent to Dave - "Congratulations on being successfully certified as a Deepworker. Now I will ask you some questions and you need to respond by pressing either the black button or the red button according to the question. Are you ready? If you understand and are ready, press the red button". Dave presses the red button.
  • Coordinating agent to Dave's earpiece - Asks a series of questions - "Was Mike professional? Press red if yes, black if no", "Was he friendly? Press red if yes, black if no", "Was Reeta professional and friendly? Press red if yes, black if no", "Did Mike have body odour or foul breath? Press red if yes and black if no", "Did Mike behave inappropriately towards you in anyway? Press red if yes, black if no" etc
  • Coordinating agent to Mike's earpiece - Asks the same set of questions about Dave.
  • Coordinating agent to Dave's earpiece - "Now walk over the stall D and observe what's happening in the room through the window." Stall D contains two people performing an activity.
  • Coordinating agent to Dave's earpiece - "Is the woman in the room standing upright on her feet? Press red if yes"
  • Coordinating agent to Dave's earpiece - "Turn around and walk to the entrance and introduce yourself as Dave, The observer to the Ram. Check your device for Ram's photo"

• And so the cycle continues. Dave is marked as qualified to receive job listings if the sum total of all the reports on him and by him are all factored in to show that he's spent an x number of hours following instructions efficiently. Once they receive sufficient amount of training hours doing this, they get to perform tasks in the real world, allowed for simple tasks in the beginning, like driving jobs, courier jobs etc.

• The whole platform is designed to minimize cognitive and behavioral load on the deepworkers and distributing effort in resource-efficient way to achieve the outcome specification most optimally. This can even operate as a decentralized autonomous organization.

• Decisions are based on crowd-sourced observations, inferences and actions of human agents that are all coordinated and channeled via the Deepwork coordinating agent, which itself uses a combination of natural language processing for procesesing language inputs/outputs, coredoc for preserving the fidelity of and mutations to the mental model associated with the action plan for outcome spec, a combination of machine learning and reasoning systems perhaps hosted on decentralized smart contracts, that guide the execution of the action plan in accordance to the dynamic mental models shaped by the human agents.

Deepwork Hardware Unit (DHU) (08/12/2025)

• When all is optimized, we'll have a situation where anyone who wishes to join our workforce will simply go to our website and fill up the forms and they will get a delivery of the welcome kit. It will be delivered by another Deepworker who would have the task of delivering the item.
• After the welcome kit is received, the package instructs them on how to get started as they open the package. They will receive a QR code that allows them to download the appropriate software on the phone to get started. And when they download the Deepwork app onto the phone, it will start off by pairing the user to the identity created during the account creation on the website. And it will guide the user through how to setup the Deepwork hardware unit.
• The Deepwork hardware unit is a wearable device that has AR capabilities and also has the ability to identify wearer through biometric authentication. The unit is wearable as VR glasses that record the video and relay it realtime into the glasses.
• The instructions for the job are relayed on the augmented display and they only display the current task at hand. And they can enable autopilot for it to just relay instructions on what to do next, while measuring their efficiency of performing the task. Once the user completes setting up the pairing of the phone to the hardware unit, the DHU simply starts instructing them on what to do step by step, one task at a time. They can optionally look into the app to check the larger context of what the job entails.
• The DHU supports voice interaction with the device, so the user can ask questions and also share information and ideas on how to execute the task better. This makes the whole plan very dynamic. And the user is incentivized to do this to optimize their efficiency score.
• The whole setup is optimized to minimize cognitive load on the Deepworker while delivering the work in the most optimal state to the customer. The Deepworker is incentivized to innovate the process and deliver services that optimize the cost of the project, this is how they build their reputation.
• Every Deepworker has a personal profile with stats on their reputation across job types. It has skill trees that allows them to progress in expertise across diverse job domains. We'll have to create a classsification system for how we broadly categorize jobs. This has to be inaccordance to how the human anatomy is structured. The job types are classified according to which organ is the primary driver of the job. A courier is all about transporting objects, so this would pertain to their feet. Because all they have to do is move their feet. So in their profile their feet will be highlighted the darkest, based on the percentage of the tasks they completed that belonged to this category. An executive who works predominantly on executive jobs will have their prefrontal cortex highlighted. This is a quick and standard way to get to know the unique strengths and experience of Deepworkers. Also we have to have a comprehensive global classification system for all possible jobs. We will have to take certain use cases to start with for our pilot program and employ those people and start delivering the work for customers in that domain and then keep diversifying.

---

Source: plan.md

Deepwork Folder Reorganization Plan

Executive Summary

Deepwork is a real-time human coordination platform concept combining AI-driven central coordination with distributed human workers wearing AR/VR guidance hardware. Currently a single 15 KB conceptual document spanning business model, technical architecture, and hardware specifications.

---

1. Content Inventory

Current State

• 1 file: Concept.md (15 KB)
• Last modified: Dec 8, 2025
• Git commits: 2
  • Dec 4: Initial concept (Overview, Concept, Core Tech)
  • Dec 8: Added DHU (Deepwork Hardware Unit) section

Content Structure

Section	Content
Overview	Platform definition, hybrid AI + human architecture
Concept	Business model, worker ecosystem, certification, revenue
Core Tech	Detailed certification flow (Dave example), technology stack
DHU	Hardware specifications, onboarding, incentive structure

---

2. Key Concepts Covered

Business Model

• Customer: Task specification for large-scale outcomes (e.g., "Clean Marina Beach in 6 hours")
• Workers: "Deepworkers" with skill-based certification and progression
• Certification Centers: Franchisable physical training centers
• Revenue: Commissions on hourly rates, job posting fees, coordination fees

Technology

• Coordination Modes: Protocol-driven manual OR AR helmet-based guidance
• Stack: Mental model coordination via Clikkin, NLP, Coredoc protocol, ML/reasoning
• Hardware: AR-capable VR glasses with biometric auth, task display, voice interaction

Worker System

• Certification: AI-generated paths, video recording, peer evaluation, crowd-sourced grading
• Progression: Skill trees, efficiency scores, innovation incentives
• Classification: Anatomically-based job categories (feet for couriers, prefrontal cortex for executives)

---

3. Questions for Organization

Q1: Document Scope

Current document mixes abstraction levels (high-level business + detailed certification flow).

Preferred organization?

• [ ] Keep as single conceptual document
• [ ] Split into separate documents by topic
• [ ] Create executive summary + detailed specs

Q2: Relationship to Clikkin

Document states Deepwork uses "Clikkin (whitelabeled platform)".

How to reflect this relationship?

• [ ] Keep Deepwork standalone
• [ ] Add Deepwork as Clikkin Space Blueprint
• [ ] Cross-reference but keep separate

Q3: Hardware vs Platform Split

DHU (hardware) section was added later and feels somewhat disconnected.

How to organize hardware content?

• [ ] Keep integrated in single document
• [ ] Create separate hardware/ folder
• [ ] Create product/ folder for both platform and hardware specs

Q4: Dave Certification Example

The detailed "Dave gets certified for carpentry" narrative is ~40% of document.

How to handle this example?

• [ ] Keep inline (illustrates concepts well)
• [ ] Extract to separate case study document
• [ ] Condense and reference detailed version

---

4. Proposed Folder Structure

Option A: Minimal (single evolved document)

Deepwork/
├── overview.yml
├── Concept.md              # Current file, improved organization
└── archive/

Option B: Modular (if expanding)

Deepwork/
├── overview.yml
├── concept/
│   ├── vision.md           # High-level overview
│   ├── business_model.md   # Customer, worker, revenue
│   └── market_analysis.md  # Competitive landscape
├── platform/
│   ├── architecture.md     # Technical stack, coordination
│   ├── certification.md    # Worker onboarding flow
│   └── job_classification.md # Skill trees, anatomy-based system
├── hardware/
│   ├── dhu_specification.md # AR glasses, biometrics
│   └── onboarding_flow.md  # Hardware setup
├── examples/
│   └── dave_certification.md # Detailed case study
└── archive/

Which structure?

• [ ] Option A (minimal)
• [ ] Option B (modular)
• [ ] Other: _______________

---

5. Gaps Identified

Consider adding documentation for:

1. Financial Model - Unit economics, pricing strategy, projections
2. Regulatory Analysis - Labor laws, worker protections, gig economy regulations
3. Technology Stack Details - Specific Clikkin integration, smart contracts
4. Pilot Program - First use case, geographic focus, success metrics
5. Competitive Analysis - vs TaskRabbit, Uber Works, existing coordination platforms
6. Risk Assessment - Quality control, safety, liability

Priority for gaps?

• [ ] Add placeholders during reorganization
• [ ] Develop later
• [ ] Not needed yet

---

6. Cross-References Needed

Deepwork connects to:

• Clikkin: Platform infrastructure
• Coredoc: Mental model preservation protocol
• SoSA: Similar distributed coordination concept
• Attentional Economy: Worker attention management

Add explicit cross-references?

• [ ] Yes
• [ ] No

---

7. Files Recommended for Archive

None currently (single file, actively relevant).

---

8. Your Responses

Q1 (Document Scope): 
Q2 (Clikkin Relationship): 
Q3 (Hardware Split): 
Q4 (Dave Example): 

Folder Structure: 
Gaps Priority: 
Cross-References:

---

Generated: 2026-01-22