Skip to content

I build marketing functions that work, and design where AI fits in the system.

Hendry Soong

CMO at getAbstract · Ex-CMO at GMS · AI Marketing Leader · AI Marketing Operator

15+ years building marketing teams and functions. Three startups I joined were later acquired. At GMS, I built the digital, content and product marketing function from scratch and led marketing communications to support international expansion across multiple markets, during a period where revenue grew from $100M to $500M.

Contact

Your Advantage isn’t the Tool. It’s the System.

The marketing leaders who win in the AI age aren’t collecting tools or mastering prompts. They’re not building a “Pile of Parts”. They think in systems, and orchestrate AI toward strategic outcomes.

FW:CTX-8E

AI marketing framework · 8 engines · 1 governed loop

Ω/Π System Orchestrate · Replicate

00→07 Closed loop Data → Listen → Signal → Create → Distribute → Convert → Measure → Iterate

00Data

makes usable inputs available

builtsample open

01Listen

detects market and customer movement

partialsample open

02Signal

holds and scores what matters

builtsample open

03Create

produces governed content artifacts

builtopen

04Distribute

routes approved work to channels

built (delivery only)not shown

Runs inside enterprise work.

05Convert

captures and qualifies response

designednot shown

There is nothing to show, because nothing was built.

06Measure

evaluates outcomes against fixed signals

built · partial (outcomes)not shown

Runs inside enterprise work.

07Iterate

proposes evidence-led Context updates

partialnot shown

Runs inside enterprise work.

−1 Context voice · ICP · messaging · governing rules

Shared law and memory

read by every engine; updated only through governed change

ΩOrchestrate

chains bounded workflows through contracts

builtsample open

ΠReplicate

ports the architecture to another brand

built · none (multi-tenant)sample open

builtpartialdesignedfine grey dots: read by every engine

Framework

Thoughts on AI Marketing

Marketing in the Agentic Age

  1. 01

    AI Marketing Framework: Eight Engines in a Governed Marketing Loop

    A governed AI marketing architecture with eight engines, shared Context, explicit contracts, validators, and human gates.

  2. 02

    Could AI Replace An Entire Marketing Team?

    88% of organisations use AI, yet martech utilisation sits at 49%. This isn’t a technology problem. It’s an architecture problem.

  3. 03

    The AI Marketing Measurement Problem: What Broke, What’s Emerging, and What to Test in 2026

    Traditional marketing measurement is breaking as zero-click searches hit 58% and AI reshapes discovery. Here are the metrics to test in 2026.

  4. 04

    Skills Benchmark Report 007: The Challenger Failed the Test We Wrote Before Building It

    A cold-email challenger met one of four checks written before it was built. It moved supplied numbers that the automated check could not see.

  5. 05

    Skills Benchmark Report 006: The Rubric Could Not See a Made-Up Name

    A cold email invented a podcast name and 37 published rules could not tell. A rule that reads the brief finds invented names in 14 of 90 counted emails.

Articles by profile56 published articles
  • operator-log 27
  • thought-leadership 17
  • definition 9
  • how-to 3
Articles by topic
  1. ai-marketing-operations34
  2. ai-marketing31
  3. ai-marketing-strategy17
  4. system-architecture12
  5. llm-validation11
  6. ai-marketing-governance5
  7. ai-marketing-agents5
  8. ai-demand-generation4
  9. context-engineering4
  10. ai-search-visibility4
  11. ai-marketing-ads3
  12. ai-content-marketing2
  13. ai-marketing-creative1
  14. visual-perception1

Explore

LOG:5 · 2026:Q3

AI Marketing Operator Logs

Public documentation of building AI Marketing Systems. What I learned.

Every iteration, failure, and architecture decision from building AI marketing capabilities in production. Timestamped, versioned, unedited.

Operator-log entries by month722 entries, Dec 2025 to Aug 2026

Dated entries on the operator-log pages, counted 9 November 2026.

Entries by capability and month1 to 910 to 2425 to 4445 to 60
capabilityDec25Jan26Feb26Mar26Apr26May26Jun26Jul26Aug26all

System

governance000000005656
orchestrate0000000236083
data verification00000106312370
data measurement0000046272158
data pipelines and sources00000914361271
gtm verification and sources0000000422769
signal0000000381048
gtm measurement0000000223355
system architecture00102140154081
deep dives0321100108

Create

create-articles8292411012268
create-images011313041326
create-compiler0041100006
create-social0100000001

Listen

listen-competitors011000100012

Replicate proof

create portability proof09001000010

Rows are the operator-log capabilities; the last column is each row’s total.

  1. July 2026release
  2. April 2026

    Build Log #1: Content System

    SystemBuild Log Series · v1.6.0

    release
  3. March 2026release
  4. February 2026release
  5. February 2026

    The Engine Split: Context Window Survival at 84K Tokens

    Create EngineCreate-Articles v7.9.2 + Create-Images v2.0.0

    release

Operator Logs

A skill claim without a measurement is marketing.

AI Marketing Skills, Measured

The AI Marketing Skills Benchmark exists to publish the measurements, including the ones that refuse to produce a number.

The seven reports in timefilled: published · hollow: updated
27 Aug 20268 Oct 2026
  1. 001
  2. 002
  3. 003
  4. 004
  5. 005
  6. 006
  7. 007
  1. 007

    Skills Benchmark Report 007: The Challenger Failed the Test We Wrote Before Building It

    A cold-email challenger met one of four checks written before it was built. It moved supplied numbers that the automated check could not see.

    Published Updated

  2. 006

    Skills Benchmark Report 006: The Rubric Could Not See a Made-Up Name

    A cold email invented a podcast name and 37 published rules could not tell. A rule that reads the brief finds invented names in 14 of 90 counted emails.

    Published Updated

  3. 005

    Skills Benchmark Report 005: The Rules Are a Floor, Not a Ranking

    Eleven arms, 99 LinkedIn posts, 14 published rules: the whole field lands inside a two-rule spread. A bare model with no skill scores 10 of 14; the best is 11.

    Published Updated

  4. 004

    Skills Benchmark Report 004: We Stopped Waiting for a Referee

    Thirteen judge seats measured, zero qualified, and the ensemble scored worse than its best member. So the benchmark switched to rules the field publishes.

    Published Updated

  5. 003

    Skills Benchmark Report 003: The Referee Exam Was Ranking Judges Backwards

    A control run with a known right answer showed the AI judge exam measured consistency, not correctness. 432 blind calls. The exam is retired.

    Published Updated

  6. 002

    Skills Benchmark Report 002: The Judge Picked Whichever Post Came First

    A judge model passed the exam three others failed, produced the first grades, and then the transcript audit found it chose whichever post came first, 90 times out of 90.

    Published Updated

  7. 001

    Skills Benchmark Report 001: The Audit That Refused Both Judges

    Hendry Soong’s AI Marketing Skills Benchmark ran its first 0.1-L1 cycle: 99 generations, zero grades. Both local judges failed a pre-registered stability audit.

    Published Updated

Skills

Proof, Not Promises

Thank you for your kind words

Hendry is what I would call an absolute marketing legend, with knowledge and experience covering many different areas.
Raj D., Chief Revenue Officer at GMS
Hendry is a strategic thinker with excellent communication skills. He grew our user base from 0 to 150k in a very short amount of time.
Maria J., CEO and Co-Founder at meetinvest
Hendry consistently demonstrated exceptional leadership and strategic marketing skills with a results-driven approach.
Olha V., Chief Growth Officer at GMS
Hendry shows great commitment, a high level of flexibility and very good analytical skills. His performance was truly outstanding.
Andrea G., CEO and Chairman at 42matters
Contact