Sanuj Krishnan Bengaluru • Work Engineering

Work Engineering • Evidence over vibes

I design AI-native organizations. Work Engineering is the spec layer.

Most companies hire AI talent. Few make work machine-readable enough for humans and agents to share it. Work Engineering specifies the work — it does not run it. Evidence over vibes. For VCs, builders, tech leaders, senior HR, and investors.

Live: Work Engineering Spec on GitHub

AI-native org Work Engineering Spec layer, not executor Evidence over vibes Human-AI org design

Now 2024 →

Co-Founder • Stealth

HIRING

Work Engineering

Zero-to-one. Specifying AI-native orgs so work is machine-readable, verifiable, and allocatable. Bengaluru.

Founding AI Engineer — Our #1 hire

2-4 years in startups, LLMs shipped to production. You ship in code, not just slides. DM me on LinkedIn.

Open live spec Apply via DM

Current Operator Role

Head HRBP @ Trianz

Services-as-Software transformation. Delayed management, redeployed talent into AI platform roles (Concierto, Avrio, Agentic AI).

— Role Maturity Framework: 16 families, 80+ jobs, benchmarked vs Amazon/Microsoft
— Co-drove AI-Native SDLC with CTO: 8 parallel PODs, Cursor + Claude Code + Codex

Builder Proof I ship in code, not just slides

work-engineering Spine

FastAPI • Postgres • React • Spec API

Specification layer between enterprise intent and execution. Work Units are independently accountable contracts. Execution systems (humans, agents, RPA, BPO) consume the spec — this product does not run the work. Live wedge: HR Offer Desk census, guest walk, 95 stated hours vs 61.8 we will defend.

Primitive

Work Unit

Stance

Spec-only

Hours

95 vs 61.8

Open Live App View Source

# spec_api.py — not an executor

def check(work_unit, evidence_ref):

if not evidence_ref:

return denied("evidence_ref required")

return allowed

stated, defended = 95, 61.8

# never blend the two numbers

▶ Guest walk: Offer Desk census

▶ Claim: spec layer, not execution

workforce-scenario-engine

Python • Monte Carlo • Streamlit

Adjacent proof, not the spine. Monte Carlo for workforce planning: hiring, attrition, AI absorption, role maturity, revenue-per-employee. Built for CHROs who think in product.

Sims

10k+

Scenarios

Parallel

Output

$/FTE

Open Live App View Source

# monte_carlo.py

def simulate_workforce(n=10000):

attrition = np.random.normal(μ, σ, n)

ai_absorption = beta(α, β, n) # 0→1

role_maturity = map_framework(16, 80+)

rev_per_fte = f(hiring, attrition, ai_absorption)

return rev_per_fte.percentile([10,50,90])

▶ 90th percentile: $90K / employee

▶ Baseline: $70K → +30% in 18mo

People Ledger

Solidity • ZK Proofs • AI Agents

Decentralized professional identity protocol. Employers cryptographically sign credentials. Employees own them. AI agents match encrypted data to jobs without revealing private ratings.

Tabs

6 Live

Layers

Trust+ZK+AI

Stack

Solidity

Open Live App View Source

# people_ledger.sol

contract PeopleLedger {

function issueCredential(bytes32 hash)

// Employer signs, employee owns

function verifyZKProof(Proof calldata p)

// Prove threshold without revealing

}

▶ AI Match: 6/6 requirements verified

▶ ZK Proof: Rating hidden ✓

Architecture studies Public-signal reconstructions. Tradeoffs named, not sold.

Atlan architecture study

Contextual intelligence • P = f(I, C)

Technical study of Atlan as an institutional brain, not a catalog with AI bolted on. Six-layer reconstruction from public UI, docs, and Gartner signals. Written for CEO / CTO / CIO / CHRO / VC.

Thesis

Context moat

Method

OSINT

Lens

C-suite

Read the study

Ontora architecture study

Interview-to-map • YC P26

How a three-person team built an AI-native process discovery engine. Six layers from campaign orchestration to ontology to human-in-the-loop. Quantification gaps and echo-chamber risk named in the open.

Thesis

Process graph

Method

OSINT

Ask

Ground truth

Read the study

Operator Proof

Revenue / Employee

+30%

$70K → $90K • 18 mo

Attrition

56%→26%

Amazon • systemic fix

Hires Scaled

4000+

People programs architected

Sr Leaders Assimilated

200+ L6+

Bar Raiser program

E-comm Org Designed

$1.5B

Carrefour MENA • 0→1

Writing Nov 2026

Essay

AI didn’t create a productivity problem. It exposed a redesign problem.

Five trade-offs every leader is making about AI, and what HR should own now.

Evidence review

The task is the unit: what the science says about redesigning work for AI

Sixty-plus sources, the open questions, and an eight-step redesign method.

Timeline

2024 — Now Co-Founder, Stealth • Work Engineering

Spec layer for AI-native orgs. Work Units, evidence gates, declared vs observed. Head HRBP @ Trianz parallel.

2023 — 2024 Head HRBP @ Trianz

Services-as-Software, Concierto/Avrio/Agentic AI roles, Role Maturity Framework (16 families, 80+ jobs), AI-Native SDLC with 8 PODs.

2021 — 2022 Carrefour MENA, Majid Al Futtaim

Designed org for $1.5B e-commerce business. Built D&I Council from zero.

2020 — 2021 Swvl • Built People from scratch across 6 countries

Restructured geography → customer-segment, accelerating 3 country launches by 6 months.

2012 — 2020 • 8+ yrs Amazon / AWS • Amazon Bar Raiser • AWS Certified

Architected people programs scaling to 4000+ hires, drove attrition 56%→26%, built Sr Leader Assimilation for 200+ L6+ leaders.

2010 — 2012 XLRI Jamshedpur • MBA

Human Resources & Organizational Design foundation.

What I do

Work Engineering

Specify work so humans and agents can share it. Work Units, Spec API, evidence gates. Not an execution engine.

AI-native org design

Delay management, create AI platform roles, design human-AI interfaces. Org that compounds with AI, not fights it.

↗

Evidence over vibes

Declared hours next to defended hours. Architecture studies from public signals. Ship the spec, name what is not built.

Connect

Bengaluru • VCs, builders, tech leaders, senior HR, investors. If you think in specs and ship in code, DM.

linkedin.com/in/sanuj-krishnan/ github.com/sanoojcools