James PustorinoApplied AI Deployment Manager, M&A TaxPwC · CPA, New York

Applied AI and agent systems for M&A tax workpapers.

James is a CPA at PwC, where he has been Manager, M&A Tax since January 2022 and has worked in applied AI since December 2022. He took Harvey, ChatGPT Enterprise, ToltIQ, and Copilot from pilot to daily use, built a tax diligence agent system that writes the request list and scope of work from the data room, and is now building a tax structuring agent system around a structure drawing that updates. Agents draft, and a named tax professional reviews and signs off on every agent output.

Email
  • 3,000+practitioners using Harvey, ChatGPT Enterprise, ToltIQ, and Copilot at 80%+ daily use
  • 2internal agent systems: diligence built, structuring building
  • 50+M&A transactions advised, $1B to $25B

What James does

He got PwC's M&A Tax practice using four AI tools on most workdays. He also builds diligence and structuring agents where a named reviewer signs off on every conclusion.

  1. Daily use

    Took Harvey, ChatGPT Enterprise, ToltIQ, and Copilot from pilot to daily use, with 40+ sessions from 2023 to 2026.

    3,000+practitioners at 80%+ daily use

  2. Tax workpaper agents

    Built a diligence system that writes the request list and scope of work from the data room. Building a structuring system where a box is a corporation, a triangle is a partnership, and an edge is the legal relationship. Written in Claude Code, Codex, and Cursor.

    2internal agent systems: diligence built, structuring building

  3. Review and sign-off

    Agents draft. A named tax professional reviews and signs off on every agent output.

    9years in tax at PwC

From deal advisory to applied AI

Nine years at PwC: fixed assets and M&A tax advisory first, then applied AI for the M&A Tax practice since late 2022.

  1. Present

    Applied AI Deployment Manager, M&A Tax

    PwCNew York, NY · December 2022 to Present

    Formal title Manager, M&A Tax since January 2022. Applied AI since December 2022. Took Harvey, ChatGPT Enterprise, ToltIQ, and Copilot to 3,000+ practitioners at 80%+ daily use. Built a tax diligence agent system that writes the request list and scope of work from the data room. Building a tax structuring agent system on an updating structure drawing. Agents draft. A named tax professional reviews and signs off on every agent output.

    • 3,000+practitioners using them
    • 80%+daily use
    • 40+training sessions
    • Took Harvey, ChatGPT Enterprise, ToltIQ, and Copilot across the M&A Tax practice from pilot to daily use.
    • Ran 40+ sessions from 2023 to 2026 and set up team champions so practitioners kept the tools in daily work.
    • Built an internal tax diligence agent system that writes the request list and scope of work from the data room, with product and engineering. Building a tax structuring agent system where a box is a corporation, a triangle is a partnership, and an edge is the legal relationship.
    • Agents draft. A named tax professional reviews and signs off on every agent output. Advises clients on AI adoption.

    Deal structuring workstream

    Illustrative acquisition structure chartA buyer owns a holding company, which owns 100 percent of the target. The target owns two corporations, drawn as boxes, and an LLC taxed as a partnership, drawn as a triangle.100%BuyerHoldCoTargetSub ALLCSub B

    CorporationTaxed as a partnership

    Agent drafts, thenReviewer signs off

    Illustrative structure. No client or deal.
  2. 2022

    Senior Tax Associate, M&A Tax

    PwCNew York, NY · January 2020 to December 2021

    Cross-border M&A tax due diligence and structuring for private equity and corporate buyers.

    • 50+M&A transactions advised
    • $1B to $25Btransaction range
    • Ran diligence and structuring workstreams for private equity and corporate transactions, from first review through post-signing steps.
    • Prepared detailed tax due diligence reports covering a target company's tax history, and modeled prospective cash tax impacts and potential leakage at exit.
    • Structured tax and legal transaction steps for buyer tax and business goals, researching statutes and case law to substantiate key positions.
  3. 2020

    Tax Associate → Senior Tax Associate, Fixed Assets

    PwCNew York, NY · June 2017 to December 2019

    Fixed asset tax planning and analysis: cost segregation, repair studies, TCJA, and depreciation ERP. The Alteryx pipeline work started here and later seeded a $5M service line.

    • $5MAlteryx service line seeded
    • Performed tax planning and analysis to deliver client tax benefits through fixed asset studies.
    • Built Alteryx pipelines for cost segregation, repair studies, TCJA models for utility clients, and depreciation ERP onboarding. Those models seeded a $5M service line and became firm-wide templates.
    • Ran cost segregation and depreciation analyses for client fixed-asset portfolios, including repair expense studies.

Tools and credentials

Rust, Python, and TypeScript for what he builds; Harvey, ChatGPT Enterprise, ToltIQ, and Copilot for PwC's M&A Tax practice. Tap any of those four to see what it actually does.

  • Languages & Interfaces

    What his agents and MCP servers are written in

    TypeScriptPythonRustMCP servers

  • Build Environments

    Where the code gets written, usually with an agent alongside

    CursorClaude CodeCodex

  • Enterprise AI in daily use

    Taken from pilot to daily use across PwC's M&A Tax practice

  • Domain

    Where he started, and still works

    AlteryxM&A tax

Credentials

  • CPANew York license, active since 2019
  • Hofstra UniversityMS Taxation and BBA Accounting, 2012 to 2016
  • Eagle ScoutBoy Scouts of America, 2012

Independent projects

Built on his own time. Every repo is public, runs, and says plainly what it can't do yet.

Python and Rust calculation are in place. Complete live Excel roundtrip acceptance remains outstanding.

Python / Rust

Workbook Forge

An Excel SDK for applications and agents.

Turns supported spreadsheet formulas and workbook structure into building blocks that software can inspect and use. Python and Rust each have their own implementation, with shared tests that compare their behavior.

View on GitHub (opens in a new tab)Explore the primitives (opens in a new tab)
Extract
Inspect supported workbook data, formulas, and their dependencies.
Compose
Use spreadsheet calculations in Python and Rust programs without Excel.
Produce
Build editable Excel workbooks with formulas, formatting, and input rules.

Live today: PTY lifecycle and governed-task producers. Still forming: generalized role contracts.

Rust

IMPULSE-rs

A local control plane for terminal-native coding agents.

It places Claude Code, Codex, and its own runtime, Ion, in explicit workspaces and supervises their processes. It records the files and tools each process touched. A task counts as accepted only when the operator accepts it against current passing evidence. Worker claims, verifier results, supervisor verdicts, and operator decisions stay separate records.

View on GitHub (opens in a new tab)

The desktop drives a real OS webview. The headless daemon cannot navigate or execute browser tools.

Rust / Tauri / React

NeuroBrowser

AI-native desktop browser.

NeuroBrowser is a Rust library with a Tauri and React shell. The desktop drives a real OS webview through tools gated by an action policy. The Rust library also supplies an HTTP scraper.

View on GitHub (opens in a new tab)

Compiling from objectives

The bet underneath every project he builds.

Today's compilers turn C into machine code. The next ones will take a stated objective, such as "given this spreadsheet, produce a tax schedule," and turn it into a working program, checked against constraints the author sets. The programmer's job moves from writing steps to writing the checks. In tax, the checks are the hard part: the rules, the deal structure, the edge cases.

What this changes in practice

Agents replace workflows
The agent works out the steps from a stated end state. The test suite replaces the flowchart.
Tests become the spec
When code is generated, the durable artifact is the test that says whether it worked. Workbook Forge has two independent calculators, one in Python and one in Rust, and one shared set of test data that both must match.
Domain knowledge is the bottleneck
Writing the code is no longer the hard part. Knowing what the program should do is: the tax rules, the deal structure, the edge cases. Nine years in tax at PwC, with M&A Tax since 2020, is the input.

Ask the career guide

Ask about James's career by text or voice. It answers only from his real record and points you to him for anything else.

  • Reads the career record, and stops at the edge of it.
  • Won't share private or client details.
  • Live ElevenLabs voice, started only with your consent.

Try a starter