I Help Corporate Professionals Gain Career Clarity & Fulfillment | Founder @Amplify Your Home | Global AI Tax Copilot Strategist @EY

Don't be fooled by my appearance. Quiet, Asian, tiny girl.

In the last decade, I've...

Successfully navigated the corporate giants: Goldman Sachs, PwC, Deloitte and EY.

Built communities, tax and financial models, and automations. (Excel wizard here)

Empowered employees to own their damn careers,

Challenge their norms to do something special.

That's when Problems become Opportunities,

Strategies into Actions.

You are not Alone.

Ready to redefine your career? Let's make it happen.

Mary Fung

McKinsey just dropped their State of AI 2025 report. The numbers are brutal.

88% of organizations say they're using AI. But only 6% qualify as 'high performers' actually seeing enterprise-level EBIT impact.

Let that sink in. 88% using, 6% winning.

What separates the 6%?

- They're 3x more likely to fundamentally redesign workflows (not just bolt AI onto existing processes)
- They spend 20%+ of digital budgets on AI
- Their senior leaders actively role model AI usage
- They've built actual data infrastructure, not just demos

Meanwhile, 64% of companies are still stuck in 'experimentation' or 'pilot' mode. Two years of pilots. Two years of POCs that never graduate.

The gap between AI leaders and everyone else is widening. Not because of better models - because of better execution.

If your AI strategy is 'let's try some use cases and see what sticks,' you're in the 94%.

That's not a strategy. That's hoping.

Own Your Damn Career | Founder @Amplify Your Home | Head of Synthetic Data & Simulations @EY

The 'AI wrapper' startups are dying. The market is finally figuring it out.

In 2023-2024, you could raise millions by putting a nice UI on top of GPT-4 and calling it a product. Those days are over.

OpenAI keeps shipping features that kill wrapper startups overnight.

Custom GPTs? Dead.

ChatGPT plugins? Dead.

The integrations announced last week? Another wave of startups just became features.

The surviving AI companies all have something the wrappers didn't:
- Proprietary data moats
- Deep workflow integration
- Vertical-specific fine-tuning
- Or actual ML teams building differentiated models

'We make ChatGPT easier to use' was never a defensible business. It just took 2 years for the market to figure that out.

RIP to the AI wrappers. 2023-2025. You will not be missed.

TCS just paid $700M for Coastal Cloud, a Salesforce consultancy. Becoming a top-5 Salesforce partner globally.

Why should you care about a consulting M&A deal?

Because it signals where the big players think enterprise AI is actually going.

TCS COO said the quiet part out loud: 'advancing our AI-led transformation agenda' through Salesforce expertise.

Not building custom AI. Not launching a foundation model. Embedding AI into the platforms enterprises already use.

This is the enterprise AI playbook that actually works:

1. Meet customers where they are (existing platforms like Salesforce)
2. Integrate AI into workflows that already exist
3. Don't ask enterprises to rip and replace - enhance what they have

The AI startups trying to replace Salesforce should be nervous. The big consultancies are betting that AI's value comes from integration, not disruption.

$700M is a lot of conviction in that thesis.

People ask why I’m okay sharing my “secrets.”

Why I teach everything I know.
Give away resources.
Tools.
Playbooks.

Because you can copy the output but
you can’t copy the process.

You won’t have my judgment.
My taste.
My reps.
My scars.

If sharing makes you afraid,
you were never the edge.

7 Lessons from My First Year Leading AI Transformation:

1. Nobody knows what they want until you show them something. Stop waiting for perfect requirements. Build a prototype and let them react.

2. Your biggest blocker isn't technology - it's the person two levels up who's afraid AI will make their team look redundant.

3. Quick wins buy you time. Deliver something useful in 30 days or leadership will move budget to whoever promises faster results.

4. The people closest to the work have the best AI use cases. Stop running executive workshops. Go sit with the analysts drowning in spreadsheets.

5. Governance isn't the enemy. The AI review board that feels like bureaucracy is actually protecting you from the lawsuit that kills the whole program.

6. You will be wrong about timelines. Multiply your estimate by 3, then add buffer for 'unknown unknowns.' You'll still probably be late.

7. Celebrate the failures publicly. When a pilot doesn't work, tell everyone why. You'll build more trust than a hundred success stories.

What Executives Get Wrong About GenAI:

> They think it's plug and play. 'Just add AI to our existing processes.' That's like adding a jet engine to a bicycle and wondering why it doesn't fly.

> They underestimate the data problem. Your AI is only as good as your data. And your data has been accumulated across 47 systems with no consistent formatting for 15 years.

> They expect immediate ROI. GenAI is infrastructure, not a feature. You don't ask 'what's the ROI on electricity?' You ask 'what can we build now that we have power?'

> They compare themselves to tech companies. Google has 10,000 ML engineers. You have Dave who took a Coursera course. Calibrate expectations accordingly.

> They skip the change management. The AI works. Your employees are terrified it will replace them. Guess which problem kills your project?

> They buy platforms before defining problems. 'We need an AI strategy' leads to expensive enterprise licenses collecting dust. Start with 'we waste 400 hours/month on X.'

Prompt engineering is just Googling with extra steps. It's not a career.

Hear me out. In 2024, we created job titles around 'prompt engineering' like it was some mystical new discipline.

Every time a new technology emerges, we create artificial scarcity around basic skills. Remember when 'social media manager' was exotic? Or when knowing HTML made you a 'webmaster'?

Prompt engineering is the same pattern. Within 2 years, it'll be assumed competency - like knowing how to use a search engine. The people building careers around it are standing on melting ice.

The real skill isn't writing prompts. It's knowing what problems to solve and having the domain expertise to evaluate the outputs. That's not prompt engineering. That's just being good at your actual job.

The “AI experiment” phase is over.

New data out of Wharton shows 75% of companies already see positive ROI from AI.

Less than 5% see negative returns.
Nearly half of business leaders use AI daily themselves.

That matches what we see on the ground.

This isn’t about prompts anymore.
It’s not “improve this email” or “summarize this doc.”

That was Phase 1.

Phase 2 has started and it’s bigger:
• “Solve this customer’s problem”
• “Handle this contract end-to-end”
• “Optimize CSAT”
• “Run this workflow”

AI is moving from accelerating steps to owning processes.

And that’s when productivity actually compounds.

The companies still stuck debating pilots and guardrails are about to feel very behind, not because AI is new, but because process always beats tools.

We’re just exiting the first act.
The second one is about systems, not shortcuts.

And once AI is embedded in how work flows - not how tasks are typed - there’s no going back.

What a time to be alive!

Why does every job posting say 'AI experience required' now?

I've been watching job boards. Finance roles want 'AI skills.' Marketing wants 'AI experience.' HR wants 'familiarity with AI tools.' Even admin assistant postings are asking for it.

But 90% of these roles don't actually need AI skills. They need people who can think critically about when NOT to use AI.

What's actually happening:
- Recruiters are adding 'AI' because everyone else is
- Hiring managers don't know what AI skills actually means
- Companies want to signal they're 'modern'
- Nobody has defined what 'AI experience' looks like

My advice to job seekers: Don't panic. 'AI experience' usually means 'can use ChatGPT without supervision.' That bar is not high.

My advice to hiring managers: Stop adding buzzwords you can't evaluate. If you can't test for it in an interview, don't put it in the job posting.

I wish I could ask Copilot about a meeting I forgot to record.

Here's the trap:

> Option 1: 'Only during the meeting' - Copilot uses temporary speech-to-text. Meeting ends, data deleted. You can't ask questions later.

> Option 2: 'During and after' - Requires saving the full transcript. Now you have retention, compliance, and legal implications.

There's no middle ground.

Missed the meeting? Too bad. Forgot to enable transcription? Too bad. Your organization restricts transcript storage for compliance? Too bad.

What I wish existed:

> A 24-hour summary window. Copilot gives you the highlights, then the raw data is deleted.

> Personal-only transcripts that don't create discoverable records.

> Retroactive summary requests. 'I was in a meeting at 2pm, can you recap?' should work.

> Graceful degradation. If full transcript isn't allowed, at least give me action items.

The 'catch up on meetings I missed' use case - the one Microsoft demos constantly - only works if someone remembered to hit record.

That's not AI. That's a tape recorder with extra steps.

taplio

The Battle

Mary Fung

Marina Levin

Followers

Posts

Engagements

Likes

Performance ScoreMary Fung's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Mary Fung

Marina Levin

Followers

Posts

Engagements

Likes

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Mary Fung's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreMary Fung's LinkedIn Performance