Heena Purohit

I’m at GitHub Universe ✨

And we’ve brought some of our best Microsoft for Startups Pegasus startups reimagining software engineering.

👋 If you’re here, come say hello.
Let’s talk about anything from AI-native software engineering to the future of work.

PS: Meet the startups shaping the new AI-native engineering stack ↓

Endor Labs → Identify and fix critical risks in complex code — whether written by humans or AI
Faros AI → Turn engineering data into actionable insights; measure productivity and ROI of coding assistants
Kubiya.ai → Agentic, context-aware DevOps copilot for every engineering team
Qodo → Ensure code integrity and quality — catch bugs, boost test coverage, enforce best practices
Roboflow → Empower every engineer with the power of computer vision in just a few lines of code

Each of them is rewriting the entire playbook of how software is built, tested, secured, deployed and maintained!

Let me know if you’d like to meet any of the incredible founders or builders behind any of these 🙌

#GitHubUniverse #agenticsoftwareengineering #GitHubCopilot

Excited to share that I’ll be speaking at Microsoft Ignite on Nov 19 in San Francisco!

Over the past year, I’ve had hundreds of conversations with enterprise leaders, helping them bridge the gap between AI hype and real-world impact.

And here’s what I've learned:
- Most incumbents are still “bolting AI on top” of legacy products. 
- It makes them maybe 10% better.

But meanwhile, 
- Startups are launching AI-native solutions that are 10x better. 
- Not just add-ons but complete rewrites of how work gets done.

Some enterprises have already realized it and are partnering with proven, enterprise-ready startups.
Others are trying to build it internally or work with consulting vendors.
Only to return months later, realizing it’s way harder than they thought.

At Ignite, I’ll be joined by amazing leaders:
Hans Yang - VP and Head of Microsoft for Startups
Surojit Chatterjee - Founder and CEO of Ema
Amee Desjourdy - CHRO, Hitachi Ltd.

Together, we'll share what’s really happening inside large enterprises, how they're winning with AI, and provide real examples and lessons learned from the front lines.

If you’re building or buying in the era of AI, this session is for you.

✨ Join us in person or watch us virtually (for free)!  
👉 Info below.

See you at #MSIgnite! 👋

#msftadvocate

2024 was the year of AI experimentation. 2025 was pilots. 2026 is when enterprises must deliver.

𝗠𝘆 𝗧𝗼𝗽 𝟭𝟬 𝗘𝗻𝘁𝗲𝗿𝗽𝗿𝗶𝘀𝗲 𝗔𝗜 𝗣𝗿𝗲𝗱𝗶𝗰𝘁𝗶𝗼𝗻𝘀 𝗳𝗼𝗿 𝟮𝟬𝟮𝟲 𝗴𝗼𝗲𝘀 𝗼𝘂𝘁 𝘁𝗼𝗺𝗼𝗿𝗿𝗼𝘄.

But before we dive into predictions, here's where we actually are now:

𝟭/ 𝗧𝗵𝗲 𝗽𝗿𝗼𝗱𝘂𝗰𝘁𝗶𝘃𝗶𝘁𝘆 𝗽𝗮𝗿𝗮𝗱𝗼𝘅 𝗶𝘀 𝗿𝗲𝗮𝗹
- Enterprises are experimenting with AI. 
- 74% teams report productivity gains, but only 11% report measurable value.
- Time saved does not automatically translate into money saved.

𝟮/ 𝗔𝗱𝗼𝗽𝘁𝗶𝗼𝗻 𝗵𝗮𝘀 𝗰𝗿𝗼𝘀𝘀𝗲𝗱 𝘁𝗵𝗲 𝘁𝗵𝗿𝗲𝘀𝗵𝗼𝗹𝗱
- AI is being adopted across organizations at unprecedented rates.
- 94% enterprises now have AI in production (up from 29% two years ago). 
- “In production” is no longer the headline. Value extraction is.

𝟯/ 𝗔𝗴𝗲𝗻𝘁𝘀 𝗮𝗿𝗲 𝗺𝗼𝘃𝗶𝗻𝗴 𝗳𝗿𝗼𝗺 𝗱𝗲𝗺𝗼𝘀 𝘁𝗼 𝗱𝗲𝗽𝗹𝗼𝘆𝗺𝗲𝗻𝘁𝘀
- For 18+ months, companies have invested heavily in agent-based pilots.
- Example use cases: software development, incident triage, ticket resolution, and workflow orchestration.
- The foundational work is done. Now comes operationalization.

𝟰/ 𝗚𝗼𝘃𝗲𝗿𝗻𝗮𝗻𝗰𝗲 𝗮𝘀 𝗮 𝘀𝗽𝗲𝗲𝗱 𝗮𝗱𝘃𝗮𝗻𝘁𝗮𝗴𝗲
- 75% executives now view AI ethics as competitive differentiation, not compliance overhead driven by regulations like the EU AI Act or evolving US policy. 
- Teams with mature governance report 52% faster time to value and 46% faster approval cycles for AI projects. 
- Good governance speeds AI up.

𝟱/ 𝗚𝗮𝗽 𝗯𝗲𝘁𝘄𝗲𝗲𝗻 𝗔𝗜 𝗿𝗲𝗮𝗱𝗶𝗻𝗲𝘀𝘀 𝗮𝗻𝗱 𝗵𝘂𝗺𝗮𝗻 𝗿𝗲𝗮𝗱𝗶𝗻𝗲𝘀𝘀
- AI capabilities are moving fast, but workforce and organizational readiness is lagging. 
- Only 11% of enterprises report having both high AI readiness and high human readiness. 62% say neither is ready. 
- The bottleneck is not the model. It is people, process and change.

𝗙𝘂𝗹𝗹 𝗹𝗶𝘀𝘁 𝗼𝗳 𝟮𝟬𝟮𝟲 𝗽𝗿𝗲𝗱𝗶𝗰𝘁𝗶𝗼𝗻𝘀 𝗰𝗼𝗺𝗲𝘀 𝗼𝘂𝘁 𝘁𝗼𝗺𝗼𝗿𝗿𝗼𝘄! It incudes more info on the above.

🤔 Are you seeing the productivity-to-value gap in organizations?

📧 I’m publishing it on Substack first. I just launched it, so the early subscribers will get it before it hits LinkedIn: https://lnkd.in/eMKUAHBw

Everyone’s racing to scale AI.
Akamai's CIO did the opposite. She slowed it down.

5 lessons from how Akamai Technologies' CIO Kate Prouty is leading the company through the AI surge 👇

1/ Move to a centralized AI strategy
- In late 2022, Akamai let all employees experiment in GenAI sandboxes.
- It fueled learning, but not scale.
- Now, all AI efforts roll up into a single, centralized AI program.

2/ Look beyond big tech and innovate with AI startups
- Yes they work with Microsoft, Google, Cisco,…
- But they also meet regularly with AI startups.
- It’s how they spot new tech early, shape their innovation roadmap, and plan investments they need to stay "AI ready".

3/ Pilot with purpose
- Every AI solution is tested in a small, measured rollout. 
- Each one with clear success metrics and strict governance.
- That's how they understand the tech and separate hype from real impact.

4/ Share what’s working (and what’s not)
- Each pilot gets a company-wide channel.
- Teams post experiments, screenshots, and lessons learned.  
- This transparency helps speed up patten recognition, and gives keeps everyone grounded in reality when trying new AI.

5/ Balance innovation with governance
- The mantra: “Encourage, don’t discourage” .
- Employees can bring any use case. IT helps them execute it securely and within cost controls.
- Even with tighter oversight, they maintain an open-door policy for new ideas.

Akamai's AI philosophy is similar to what I'm seeing at across enterprises:
AI adoption isn’t just about speed. It’s about structure.
Experiment boldly. Deploy carefully. Keep trust and accountability at the core.

👉 How is your company balancing AI innovation within guardrails? 
💬 Any other best practices you’d like to share?

🔗 Full article in comments.

----
♻️ Share with others that need to hear this.
🔔 Follow Heena Purohit for similar posts.

Huge Immigration Update: EB-1 priority dates for India jumped by ~1 year.

This is great news for the incredible, high-skilled talent coming out of India and China.

Thousands of them are now eligible to apply for their green cards, and many more should be receiving theirs in the mail.

When I look back at decisions in my life that had took a LOT of time and effort but had outsized returns, pursuing my EB1-A is one of them.

If you’re on the fence about whether you qualify, I highly recommend doing a discovery call with an attorney. You have nothing to lose.

I previously shared an open, tell-all webinar about my experience self-filing my EB-1A and the lessons I learned along the way. Linking it below.

👉 Jan 2026 Visa Bulletin: https://lnkd.in/eznixH3g

#immigration #EB1A

📢 NYC devs: The GitHub team is hosting a mid-week hangout for you tomorrow.

The GitHub Social Club is organizing an all-day, drop-in, co-work + conversation space.
No panels. No pitch decks.
Just a space for teams that actually ship things to connect and hang out.

They’ll also have people who are building Copilot, Next, Developer Productivity, GitHub Fund, and Startups teams.

If you wanna talk about coding, open source, startups or even the bug you solved at 2 AM, drop by (space permitting) or sign up for a guaranteed spot!

👉 Registration: https://lnkd.in/eq6UWvfN
📍 Venue: Bibliotheque, SoHo (NYC)

♻️ Share this with others in NYC that should know about this!

#github #coding #copilot Kevin Crosby Corey Dolik Microsoft for Startups #msftadvocate #NYCevents Microsoft for Startups

🚀 Excited to welcome NeuBird.ai to the Microsoft for Startups Pegasus Program!

They’ve built Hawkeye, the world’s first agentic AI Site Reliability Engineer (SRE).

It doesn’t just observe. It acts.
Built to autonomously diagnose incidents, find root causes, and trigger real-time remediation.

For enterprise IT and DevOps teams, that means:
- Up to 90% faster incident resolution
- Far fewer 3 a.m. alerts.
- More time for engineering that actually moves the business

Glad to have Gou Rao, Patrick Brennan, Paul Searles, Shilpi Srivastava, Justin Griffin in the program, building what's next in agentic reliability engineering.

The future of DevOps just got a lot more autonomous.

👇 If you're focused on DevOps productivity, check them out below.
or DM me for an intro.

Tom Davis Sally Ann Frank ShiSh Shridhar Kevin Magee Kevin Li-Kai Kuo Tiffany Johnson Jared Prins Bethany Cordes M12, Microsoft's Venture Fund Michael Stewart Jose Clautier Marion Desmazières

#MicrosoftForStartups #PegasusProgram #AgenticSoftwareEngineering #DevOpsTransformation

Software is becoming disposable. And that's actually a good thing.

Non-developers are now building “micro apps” the way we used to build spreadsheet templates.

𝗦𝗼𝗺𝗲 𝗲𝘅𝗮𝗺𝗽𝗹𝗲𝘀:
- I used Replit to build an app to track my intermittent fasting schedule. 
- A dining app to decide where to eat with friends
- A game to play with the family over the holidays.
- A parking ticket helper to auto-pay parking tickets.

None of these are polished products headed for the App Store.

They're software tools that are built solve one problem, then disappear when the need is gone.

𝗪𝗵𝘆 𝗶𝘀 𝘁𝗵𝗶𝘀 𝗵𝗮𝗽𝗽𝗲𝗻𝗶𝗻𝗴? 
- With vibe coding tools like Replit, Lovable, Cursor, Claude, anyone can describe an app in plain language and watch it appear.
- Zero coding required.
- And this entire process is surprisingly gratifying.

This trend started as a consumer play. 
But I'm already seeing it coming for enterprise workflows.

I don't see it disrupting major software categories yet.
Teams aren’t just going to just abandon their CRM system.

But the interesting question is what else people start building in-house.

𝗦𝗼𝗺𝗲 𝗲𝘅𝗮𝗺𝗽𝗹𝗲𝘀 𝗜 𝗰𝗮𝗻 𝘁𝗵𝗶𝗻𝗸 𝗼𝗳:
- Product teams can build rapid prototypes and perform customer discovery. 
- SalesOps teams can build custom lead scoring tools.
- HR teams can create employee onboarding trackers themselves
And more!

The gap between "I wish this existed" and "I built it" is going to shrink.

🤔 Have you vibe coded a personal app yet? Why or why not?

♻️ Repost this to help your network stay on top of AI trends
➕ Follow Heena Purohit for more

🔗 More reading in comments.

When a founder is deeply passionate about the problem they’re solving and crystal clear on their mission, you can feel it instantly.

That’s Steven Aberle, founder and CEO of Rohirrim.

He’s one of those rare AI-native founders who’s not just building fast, he’s building right.

In our upcoming LinkedIn Live, we’ll talk about:
- His founder journey from idea to scale
- Building a moat in today’s AI age
- Selling AI into regulated industries

🎬 We had so much fun prepping for this session, and I can’t wait to bring that same energy to you all live!

💡 Tune in to hear lessons every AI founder and enterprise team need to hear.

🎙️ 𝗝𝗼𝗶𝗻 𝘂𝘀 𝗟𝗜𝗩𝗘 𝘁𝗼𝗺𝗼𝗿𝗿𝗼𝘄: https://lnkd.in/e5jMfPJN
Oct 24 | 1 PM ET / 10 AM PT

PS: It’s my first LinkedIn Live! It would be great to see you there.
Bring your curiosity, your questions, and your coffee ☕️

Microsoft for Startups #LinkedInLive #AIStartups #EnterpriseAI #AIInnovation

🚀 AI isn’t just helping write code faster. It’s rewriting the entire playbook of how software is built, tested, secured, deployed and maintained.

And you can see it all firsthand at GitHub Universe 2025.

This year, I’m bringing an incredible lineup of Microsoft for Startups Pegasus startups reimagining software engineering.

Each of them is redefining a layer of the new AI-native engineering stack:
Anyscale → Run and scale every AI/ML workloads seamlessly, from laptop to cloud
Endor Labs → Identify and fix critical risks in complex code - whether written by humans or AI
Faros AI → Turn engineering data into actionable insights; Measure productivity and ROI of coding assistants
Kubiya.ai → Agentic, context-aware DevOps copilot for every engineering team
Qodo → Ensure code integrity and quality - catch bugs, boost test coverage, enforce best practices, etc.
Roboflow → Empower every engineer with the power of computer vision in just a few lines of code

These aren’t just tools.
They’re a glimpse into how software engineering is evolving:
→ from reactive to proactive
→ from manual to agentic
→ from siloed to fully autonomous

If you’re heading to #GitHubUniverse, come find me and meet the founders transforming how the world builds and runs software.

📌 Learn more about these enterprise-ready startups: https://lnkd.in/e6hNBQ8n
📌 To schedule a meeting with any of them, DM me or reach out at: mfsgtm@microsoft.com

PS: There's only a couple of seats left for my Engineering Leaders Dinner on Oct 27. Let me know if you want in!

Microsoft Developer Microsoft Events #AgenticSoftwareEngineering #EnterpriseAI #AIforBusiness

📣 Startup Teams: Enterprise AI in 2026 has a new scoreboard.

Next week, I’m co-hosting a webinar on State of Enterprise AI in 2026.

Where I’m sharing insights learned from 100s of 1:1 conversations with enterprise AI leaders.

We’ll cover:
→ What I’m hearing from enterprise buyers
→ The patterns: What’s working, what’s not
→ Where startup teams keep getting stuck
→ How to position yourself as an AI startup

Note: 𝗧𝗵𝗶𝘀 𝗶𝘀 𝗲𝘅𝗰𝗹𝘂𝘀𝗶𝘃𝗲𝗹𝘆 𝗳𝗼𝗿 𝗠𝗶𝗰𝗿𝗼𝘀𝗼𝗳𝘁 𝗳𝗼𝗿 𝗦𝘁𝗮𝗿𝘁𝘂𝗽𝘀 𝗳𝗼𝘂𝗻𝗱𝗲𝗿𝘀.
Feb 5 | 12 PM - 1 PM ET | Virtual
Comment “INVITE” and I’ll DM the registration details.

Not in the program yet? what are you waiting for 🤷🏻‍♀️
→ Join here: https://lnkd.in/esRKK3FH
→ Then sign up to join this and many other knowledge sessions we host regularly.

♻️ Know a founder who should hear this? Tag them or send this their way

The best engineering leaders aren’t “adding more AI.”
They’re designing smarter systems for people.

Here are the top 5 things enterprise leaders shared about their AI use:

𝟭/ 𝗠𝗮𝗻𝘆 𝘁𝗲𝗮𝗺𝘀 𝘂𝘀𝗲 𝗺𝘂𝗹𝘁𝗶𝗽𝗹𝗲 𝘁𝗼𝗼𝗹𝘀
- Each tool has different strengths.
- So engineers are given access to all of them; and they choose the tool they want. Sometimes it’s preference, sometimes task-specific.
- Most are great for greenfield development (starting from scratch).
- Some are bad at long-form manipulation (e.g. changing code across 500+ repos.
- Some are bad at ingesting a large codebase.
- They’ll all get better, but for now: there are preferences, styles, strengths and weaknesses.

𝟮/ 𝗖𝗼𝗱𝗶𝗻𝗴 𝗶𝘀 𝗮 𝘀𝗺𝗮𝗹𝗹 𝗽𝗮𝗿𝘁 𝗼𝗳 𝗮 𝘀𝗼𝗳𝘁𝘄𝗮𝗿𝗲 𝗱𝗲𝘃𝗲𝗹𝗼𝗽𝗲𝗿’𝘀 𝗷𝗼𝗯. 
- Actual coding might be only 10-30% of the week.
- The rest? Tasks like search, debugging, reviews, compliance, change-management
- Teams are applying AI to optimize those, too

𝟯/ 𝗧𝗵𝗲 𝗹𝗲𝗴𝗮𝗰𝘆 𝗰𝗼𝗱𝗲 𝗽𝗿𝗼𝗯𝗹𝗲𝗺 𝗶𝘀 𝗿𝗲𝗮𝗹. 
- A huge chunk of code may be written by people who’ve since left the org.
- Teams still have to understand, maintain, and even modernize it

𝟰/ 𝗔𝗻𝗱 𝗻𝗼, 𝘁𝗵𝗲 𝗮𝗻𝘀𝘄𝗲𝗿 𝗶𝘀𝗻’𝘁 “𝘁𝗵𝗿𝗼𝘄 𝗺𝗼𝗿𝗲 𝗔𝗜 𝗮𝘁 𝗶𝘁.”
- That’s how you get AI solving AI-created problems.
- Tools need to be integrated thoughtfully. Avoid tool sprawl.

𝟱/ 𝗧𝗵𝗲 𝘄𝗶𝗻𝗻𝗶𝗻𝗴 𝘁𝗲𝗮𝗺𝘀 𝗮𝗿𝗲 𝗳𝗼𝗹𝗹𝗼𝘄𝗶𝗻𝗴 𝗼𝗻𝗲 𝗼𝗿𝗱𝗲𝗿:
- People → process → tools.
- Start with workflows, ownership, and guardrails.
- Then pick the smallest tool that actually helps you drive some outcomes.

💬 𝗜𝗳 𝘆𝗼𝘂’𝗿𝗲 𝗮𝗻 𝗲𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴 𝗹𝗲𝗮𝗱𝗲𝗿 𝘁𝗵𝗶𝗻𝗸𝗶𝗻𝗴 𝗱𝗲𝗲𝗽𝗹𝘆 𝗮𝗯𝗼𝘂𝘁 𝘁𝗵𝗶𝘀… 
I’m hosting another private dinner on October 26 in SF to swap real stories on AI adoption, developer experience, and what’s actually working.
DM me if you’d like to join the table!

🤔 𝗛𝗼𝘄 𝗮𝗿𝗲 𝘆𝗼𝘂 𝗮𝗰𝘁𝘂𝗮𝗹𝗹𝘆 𝘂𝘀𝗶𝗻𝗴 𝗔𝗜 𝗶𝗻 𝘆𝗼𝘂𝗿 𝘀𝗼𝗳𝘁𝘄𝗮𝗿𝗲 𝘁𝗲𝗮𝗺𝘀 𝘁𝗼𝗱𝗮𝘆?

#softwareengineering #agenticsoftwareengineering #enterpriseAI #GitHubUniverse25

Hot take: Everyone’s talking about building agents.

Here’s why you should NOT be building an AI agent right now.

I’m seeing this pattern repeat across enterprise and startup teams:
They’re building the wrong thing, at the wrong time, for the wrong reasons.

Let’s unpack the top reasons 👇

1/ 𝗧𝗮𝘀𝗸 𝘃𝗼𝗹𝘂𝗺𝗲 𝗶𝘀 𝘁𝗼𝗼 𝗹𝗼𝘄
- Automation only pays off when the task happens *often* enough to justify build costs.
- If you’re handling 200 support tickets a month, you don’t need a $50K agent
you need better workflows, documentation, or process hygiene.
- Agents shine when scale and repetition meet friction.

2/ 𝗬𝗼𝘂𝗿 𝗱𝗮𝘁𝗮 𝗶𝘀 𝗰𝗵𝗮𝗼𝘀
- Agents need clean, connected, context-rich data to make good decisions.
- If it’s scattered across CRMs, spreadsheets, and SharePoint folders, you’re not building an agent, you’re building confusion.
- Before you automate, fix the plumbing.
- Data readiness isn’t glamorous, but it’s the foundation every agent stands on.

3/ 𝗬𝗼𝘂 𝗰𝗮𝗻’𝘁 𝗱𝗲𝗳𝗶𝗻𝗲 𝘀𝘂𝗰𝗰𝗲𝘀𝘀
- Without measurable KPIs: e.g., time saved, cost avoided, or satisfaction 
- You’ll never know if the agent’s actually working.
- AI needs KPIs just like any product.

4/ 𝗧𝗵𝗲 𝘁𝗮𝘀𝗸 𝗶𝘀𝗻’𝘁 𝘄𝗼𝗿𝘁𝗵 𝗮𝘂𝘁𝗼𝗺𝗮𝘁𝗶𝗻𝗴
- Just because something can be automated doesn’t mean it should be.
- Low-effort, low-impact tasks (like tagging an email) are quick wins but don’t transform the business.
- E.g., If a company spends months automating a report their intern did in 30 minutes. Guess who won? The intern.
- If automation doesn’t unlock time, insight, or decision-making capacity, it’s a distraction not innovation

5/ 𝗡𝗼 𝗼𝗻𝗲 𝗼𝘄𝗻𝘀 𝗶𝘁
- Agents need monitoring, tuning, debugging, retraining.
- Without a technical owner, it’ll quietly degrade until nobody uses it.
- If IT, data, and business teams can’t name who’s accountable,
your agent will drift into pilot purgatory.

𝗕𝗼𝘁𝘁𝗼𝗺 𝗹𝗶𝗻𝗲:
Yes, you can build an agent. 
But that doesn’t mean you should.

Don’t building an agent for agent’s sake. 
And focus on being agent ready first

💬 I’m curious: What’s the top blocker you’ve seen for AI agent projects?

---- 
♻️ Share with others that need to hear this.
😅 Or others who love The Office memes, too.
🔔 Follow Heena Purohit for similar posts.

📢 The Copilot feature you need to know: Announcing Copilot Podcasts ✨

Stop reading long docs.
Start listening to them like a podcast.

🎧 Copilot Podcasts automatically turn your docs, research, PRDs, GitHub issues, project briefs, or meeting notes into a podcast-style audio conversation.

It goes beyond summarizing text.
Copilot generates:
- A dynamic, two-way conversation
- Contextual insights and key takeaways
- A shareable audio experience you can consume on the go

See it in action below👇
This podcast summarizes the announcements at GitHub Universe 2025, with a focus on the most transformative one: Agent HQ!

🔗 Full GitHub Universe recap in comments.
🎧More info on how to generate your own Copilot podcast below!
💡 Follow me Heena Purohit for more on AI in the real world.

Microsoft Copilot Microsoft Developer Microsoft 365 Developer #msftadvocate #msftcopilot #githubuniverse2025

I asked 100+ AI builders about their favorite models.

Here are the best AI models, by use case:

Coding - Opus and OpenAI Codex 
Images - Nano Banana Pro 
Video - Sora-2 
Best general-purpose model - GPT 5.2
Best small model - Gemini Flash 3.0 
Planning - GPT 5.2 thinking 
Research - Gemini 3.0
Real-time - Grok 4.1

The Qwen models are a close contender for many of these. Teams are using Alibaba’s open-source models as their base for fine-tuning.

Which model is your go-to right now, and for what use case?

♻️ Repost this to help your network stay on top of AI
➕ Follow Heena Purohit for more

📢 Microsoft just launched three new AI certifications for AI literacy in the workplace!

If you want to stay relevant in the age of Copilot, AI agents, and autonomous workflows, check them out:

✨ AI Business Professional (AB-730)
For business users who want AI-powered productivity, without the technical depth: https://lnkd.in/eCgm6SW5

✨ AI Transformation Leader (AB-731)
For execs and change leaders driving AI strategy, ROI, governance, and adoption: https://lnkd.in/esmNPyqH

✨ M365 Copilot & Agent Fundamentals (AB-900)
For IT teams + admins managing identity, governance, access, security, and AI-enabled M365 environments: https://lnkd.in/eZk-gXzG

I get asked for AI course recommendations all the time.
But most options are either too technical, too high-level, or not relevant enough for real workplace skills.

These three certifications fill that gap.
They help you build, and prove, AI fluency.

Choose the one that aligns with your role and the impact you want to make!

---- 
♻️ Share with others that need to know this.
🔔 Follow Heena Purohit for similar posts.

#msftadvocate #AILiteracy #EnterpriseAI #AIatWork

🚀 AI Startups: Why You Can’t Miss Microsoft Ignite 2025

Every year, Ignite sets the tone for where enterprise AI is headed.
This year might be the most pivotal one yet.

If you’re building in AI, here are 3 reasons you’ll want to clear your calendar👇

1/ 𝗘𝗮𝗿𝗹𝘆 𝘀𝗶𝗴𝗻𝗮𝗹𝘀 𝗼𝗻 𝗔𝗜 𝗽𝗹𝗮𝘁𝗳𝗼𝗿𝗺 𝗯𝗲𝘁𝘀
Ignite is where we make some of the BIGGEST announcements.
Everything from Copilot and agents to Azure AI Foundry, cloud and security.
Use this to align your roadmap, integrate or pivot faster than anyone else.

2/ 𝗥𝗲𝗮𝗹 𝘁𝗮𝗹𝗸 𝘄𝗶𝘁𝗵 𝗲𝗻𝘁𝗲𝗿𝗽𝗿𝗶𝘀𝗲 𝗰𝘂𝘀𝘁𝗼𝗺𝗲𝗿𝘀
Hear directly from enterprise leaders.
Learn what they’re doing, what’s working, and the challenges they’re hearing. 
Validate if your product is vitamin or a painkiller.

3/ 𝗨𝗻𝗹𝗼𝗰𝗸 𝘁𝗵𝗲 𝗽𝗼𝘄𝗲𝗿 𝗼𝗳 𝘁𝗵𝗲 𝗠𝗶𝗰𝗿𝗼𝘀𝗼𝗳𝘁 𝗲𝗰𝗼𝘀𝘆𝘀𝘁𝗲𝗺
If you’re a B2B startup, I can’t think of a better distribution channel than Microsoft!
Between the Marketplace, co-sell programs, and Copilot integrations, 
you can plug into a global sales engine instead of building one from scratch.

💡I’ve shared more info on all of this + added pro tips based my experiences at last year’s Ignite in my latest blog post.

Check it out below👇

🔥 𝗔𝗿𝗲 𝘆𝗼𝘂 𝗿𝗲𝗮𝗱𝘆 𝘁𝗼 𝗜𝗴𝗻𝗶𝘁𝗲? 
Let me know if you will be there!

Microsoft Events Microsoft Developer #MSIgnite #MSIgnite2025

🚀 AI Startups: Why You Can’t Miss Microsoft Ignite 2025

Every year, Ignite sets the tone for where enterprise AI is headed.
This year might be the most pivotal one yet.

If you’re building in AI, here are 3 reasons you’ll want to clear your calendar👇

1/ 𝗘𝗮𝗿𝗹𝘆 𝘀𝗶𝗴𝗻𝗮𝗹𝘀 𝗼𝗻 𝗔𝗜 𝗽𝗹𝗮𝘁𝗳𝗼𝗿𝗺 𝗯𝗲𝘁𝘀
Ignite is where we make some of the BIGGEST announcements.
Everything from Copilot and agents to Azure AI Foundry, cloud and security.
Use this to align your roadmap, integrate or pivot faster than anyone else.

2/ 𝗥𝗲𝗮𝗹 𝘁𝗮𝗹𝗸 𝘄𝗶𝘁𝗵 𝗲𝗻𝘁𝗲𝗿𝗽𝗿𝗶𝘀𝗲 𝗰𝘂𝘀𝘁𝗼𝗺𝗲𝗿𝘀
Hear directly from enterprise leaders.
Learn what they’re doing, what’s working, and the challenges they’re hearing. 
Validate if your product is vitamin or a painkiller.

3/ 𝗨𝗻𝗹𝗼𝗰𝗸 𝘁𝗵𝗲 𝗽𝗼𝘄𝗲𝗿 𝗼𝗳 𝘁𝗵𝗲 𝗠𝗶𝗰𝗿𝗼𝘀𝗼𝗳𝘁 𝗲𝗰𝗼𝘀𝘆𝘀𝘁𝗲𝗺
If you’re a B2B startup, I can’t think of a better distribution channel than Microsoft!
Between the Marketplace, co-sell programs, and Copilot integrations, 
you can plug into a global sales engine instead of building one from scratch.

💡I’ve shared more info on all of this + added pro tips based my experiences at last year’s Ignite in my latest blog post.

Check it out below👇

🔥 𝗔𝗿𝗲 𝘆𝗼𝘂 𝗿𝗲𝗮𝗱𝘆 𝘁𝗼 𝗜𝗴𝗻𝗶𝘁𝗲? 
Let me know if you will be there!

Microsoft Events Microsoft Developer #MSIgnite #MSIgnite2025

Total predicted Al spend in 2026: $2.5 trillion.

That's a 44% jump from 2025.

The biggest drivers:
- Al infrastructure: $1.37 trillion (+42% YoY).
- Al software: $452 billion (+60% YoY).

But most Al budgets still miss the full picture.

If you're budgeting for Al projects, make sure don't just look at model costs.
Budget for:
- The software around it.
- The infrastructure (including compute, services, backups, etc.)
- The implementation costs
- Costs to deploy and maintain it
- People costs: including training, change management.

Factor all of this into your Al budget and ROI calculations.

---
♻️ Repost this to help your network stay on top of Al trends
➕ Follow Heena Purohit for more

🔗 Source: Gartner. Link in comments.

YC just dropped their Spring 2026 “Request for Startups”.

The biggest shift?
They're not asking for AI apps or agents. They're asking for AI to rebuild how work gets done: from trading desks to factory floors to government offices.

Here are 10 ideas YC wants to fund right now:

1/ Cursor for Product Managers
2/ AI-Native Hedge Funds
3/ AI-Native Agencies
4/ Stablecoin Financial Services
5/ AI for Government
5/ Modern Metal Mills
6/ AI Guidance for Physical Work
7/ Large Spatial Models LLMs
8/ Government Fraud Infrastructure
9/ Making LLMs Easy to Train

It's been interesting to see how YC's Requests for Startups are evolving.

They're now funding startups that fundamentally rewire how work gets done: whether that's a product manager deciding what to build, a hedge fund making trades, a nurse learning a new procedure, or a government office catching fraud.

This isn’t hype. It’s YC saying: the era of bolting AI onto old businesses is over. The winners will be the ones that rebuild the business from the ground up with AI.

💬 Anyone building in these spaces?

Learn more and apply: https://lnkd.in/ebX2BvZd
#startups #founders #startuptips Y Combinator

Why I’m bullish on Open Source AI:

There’s a new model release almost every week.
It’s moving so fast that I don’t talk about new models unless I think it’s category defining.

𝗕𝘂𝘁 𝗼𝗻𝗲 𝗸𝗲𝘆 𝘁𝗿𝗲𝗻𝗱 𝘁𝗵𝗮𝘁’𝘀 𝘁𝗼𝗼 𝗶𝗺𝗽𝗼𝗿𝘁𝗮𝗻𝘁 𝘁𝗼 𝘀𝗸𝗶𝗽:
The dramatic evolution of market share of top open-source models.

In late 2024, the market was highly consolidated.
DeepSeek models dominated, with over half of all open source token usage

By mid-2025, the picture completely changed:
The market broadened and deepened.

New entrants like Qwen's models, Minimax's M2, MoonshotAl's Kimi K2, and OpenAl's GPT-OSS series all came in and grew rapidly.
And they exploded in adoption, capturing meaningful market share.
They often reached production-scale adoption within weeks of launch.

This shift tells us two things:

1. 𝗧𝗵𝗲 𝗯𝗲𝘀𝘁 𝗔𝗜 𝘁𝗲𝗮𝗺𝘀 𝗺𝗼𝘃𝗲 𝗳𝗮𝘀𝘁. 
- I've heard from many AI teams that they test new models the week they launch.
- They quickly adopt models that outperform on performance, quality, latency, cost.

2. 𝗢𝗽𝗲𝗻-𝘀𝗼𝘂𝗿𝗰𝗲 𝗲𝗰𝗼𝘀𝘆𝘀𝘁𝗲𝗺𝘀 𝗮𝗿𝗲 𝘀𝗰𝗮𝗹𝗶𝗻𝗴 𝗳𝗮𝘀𝘁𝗲𝗿 𝘁𝗵𝗮𝗻 𝗲𝘃𝗲𝗿.
- Open-source models are now competitive with frontier models for real-world use cases.
- They offer customization, flexibility, and cost-efficiency that many teams need.

The demand is there.
Supply is growing. And faster than ever.

If you're building AI solutions: Make sure you don't ignore the open-source ecosystem.

💭 I'm curious: Are you betting more on open-source or closed models?

Image Source: OpenRouter, Inc.
More info in comments.

#opensource #AI #EnterpriseAI

The first large-scale study of AI agents in production was just published.

Most teams think agentic AI is all about complexity. 
This research shows the opposite.

Researchers surveyed 306 practitioners + conducted 20 in-depth case studies across 26 domains to learn how teams are really really building agents and deploying them in production.

7 insights every AI team should know:

𝟭. 𝗦𝗶𝗺𝗽𝗹𝗲 𝗮𝗴𝗲𝗻𝘁𝘀 𝗼𝘂𝘁𝗽𝗲𝗿𝗳𝗼𝗿𝗺 𝗳𝗮𝗻𝗰𝘆 𝗼𝗻𝗲𝘀
70% rely on prompting off-the-shelf models (GPT/Claude).
68% cap autonomy at ≤10 steps before needing human review.
47% complete fewer than 5 steps.
→ Complexity is not the bottleneck. Reliability is.

𝟮. 𝗛𝘂𝗺𝗮𝗻 𝗼𝘃𝗲𝗿𝘀𝗶𝗴𝗵𝘁 𝗶𝘀 𝗮 𝗻𝗼𝗻-𝗻𝗲𝗴𝗼𝘁𝗶𝗮𝗯𝗹𝗲
74% use human evaluation as their primary safeguard.
LLM-as-a-judge is growing (52%), but never replaces humans.
→ Trust = human + model, not model alone.

𝟯. 𝗥𝗲𝗹𝗶𝗮𝗯𝗶𝗹𝗶𝘁𝘆 𝗶𝘀 𝘁𝗵𝗲 𝘁𝗼𝗽 𝗯𝗹𝗼𝗰𝗸𝗲𝗿
Reliability, robustness, and correctness are the biggest blockers.
Teams consistently trade autonomy for predictability.
They use bounded workflows, sandboxing, and read-only modes.
→ Teams don’t want “agents that roam.” They want “agents that don’t break.”

𝟰. 𝗣𝗿𝗼𝗱𝘂𝗰𝘁𝗶𝘃𝗶𝘁𝘆, 𝗻𝗼𝘁 𝗻𝗼𝘃𝗲𝗹𝘁𝘆, 𝗱𝗿𝗶𝘃𝗲𝘀 𝗮𝗱𝗼𝗽𝘁𝗶𝗼𝗻
Main reason for agent deployment: efficiency + human-hour reduction (73%).
Risk mitigation and innovation rank much lower.
→ Agents succeed when they replace manual, repetitive tasks.

𝟱. 𝗘𝘃𝗮𝗹𝘂𝗮𝘁𝗶𝗼𝗻 𝗹𝗮𝗴𝘀 𝗯𝗲𝗵𝗶𝗻𝗱 𝗱𝗲𝗽𝗹𝗼𝘆𝗺𝗲𝗻𝘁.
Most teams don’t have standardized benchmarks.
They rely on: A/B testing, domain expert review, or user feedback loops.
→ Your evaluation pipeline can be a moat.

𝟲. 𝗔𝗴𝗲𝗻𝘁 𝗙𝗿𝗮𝗺𝗲𝘄𝗼𝗿𝗸𝘀 𝗮𝗿𝗲 𝗳𝗿𝗮𝗴𝗺𝗲𝗻𝘁𝗲𝗱
Per the survey, ~60% use 3P frameworks like LangChain/LangGraph…
…but 85% of interviewed teams chose custom implementations from scratch.
Core agent loops are straightforward to implement with direct API calls.
→ Mature teams prefer minimal dependency bloat and abstraction layers for greater control, security, and simplicity.

𝟳. 𝗠𝘆 𝘁𝗮𝗸𝗲𝗮𝘄𝗮𝘆 𝗳𝗼𝗿 𝗔𝗜 𝘁𝗲𝗮𝗺𝘀
→ Choose deliberate simplicity, not sophisticated autonomy.
→ Start narrow, build reliability, then increase autonomy.
→ Design for evaluation: bake in human and automated checks early.
→ Measure impact through productivity gains and outcomes, not model scores.
→ Prioritize control + transparency over complexity.

💭 How are you building AI agents? Does this align with what you're seeing?
🔗 Link to paper in comments

---- 
♻️ Share with others that need to hear this.
🔔 Follow Heena Purohit for similar posts.

You might think AI is making you more productive.
But is it, really?

New research from an 8-month study found something counterintuitive: AI didn't reduce work. It intensified it.

Here's what happened when employees were adopting AI:

𝟭. 𝗧𝗮𝘀𝗸 𝗲𝘅𝗽𝗮𝗻𝘀𝗶𝗼𝗻
They worked faster, took on broader tasks, and extended work into more hours.
→ PMs and designers started writing code
→ Researchers took on engineering work
→ Engineers spent more time reviewing and correcting AI-generated code.

𝟮. 𝗕𝗹𝘂𝗿𝗿𝗲𝗱 𝗯𝗼𝘂𝗻𝗱𝗮𝗿𝗶𝗲𝘀
"Quick last prompts" before stepping away became the norm.
Because prompting feels like chatting, not working.
This meant fewer real pauses. Downtime stopped restoring people.

𝟯. 𝗠𝗼𝗿𝗲 𝗺𝘂𝗹𝘁𝗶𝘁𝗮𝘀𝗸𝗶𝗻𝗴
AI enables parallel threads, multiple agents, multiple drafts, revived backlog.
It feels like momentum. But it's actually constant context-switching.

It's also important to call out nobody asked the employees to do more work.
AI just made "doing more" feel possible.

This is the hidden cost of AI productivity gains: What looks like a win today can quietly become a workload creep and cognitive strain tomorrow.

𝗜𝗳 𝘆𝗼𝘂'𝗿𝗲 𝘂𝘀𝗶𝗻𝗴 𝗔𝗜 𝘁𝗼𝗱𝗮𝘆, 𝗿𝗲𝗺𝗲𝗺𝗯𝗲𝗿 𝘁𝗼:
→ Use AI to augment your thinking, not replace it
→ Protect your pauses. They're not laziness. They're how your brain consolidates what you've learned
→ Prioritize human connections

The real promise of AI isn't that you can do more. It's that you can do better.

💭 I'm curious: Have you seen AI reduce work, or quietly expand it?

---
♻️ Repost this to help your network lead with AI
➕ Follow Heena Purohit for more

🔗 Source: Harvard Business Review. Link and more info in comments.

Stop saying “95% of AI projects fail.”

That stat from the MIT study is misleading.
And it’s holding AI teams back.

Enterprises are actively experimenting with AI.
And making decisions on what to scale.
If you’re standing still, you’re already behind.

Here are 5 recent reports that show how enterprise are actually adopting AI 👇

1/ State of AI in 2025 - McKinsey & Company
https://lnkd.in/eDFn7UqN

2/ State of Enterprise AI - OpenAI
https://lnkd.in/e7Ns9heX

3/ How "Frontier Firms" are transforming business - Microsoft
https://lnkd.in/e3U-Nkff

4/ Enterprise AI Adoption Patterns - Anthropic Economic Index
https://lnkd.in/e_HxExKf

5/ State of Generative AI in the Enterprise - Menlo Ventures
https://lnkd.in/efURUqsm

𝗧𝗵𝗲 𝗿𝗲𝗮𝗹𝗶𝘁𝘆:
Companies are deploying AI and seeing real value.
Yes, some AI initiatives fail. Teams learn. That's how progress happens.

If you’re preparing your organization for AI in 2026, start by understanding what already works.

📌 Save this reading list for later.
You’ll want to come back to it during roadmap planning.

🤔 Curious: Any other resources you’d recommend?

PS: I’ve written earlier about how the MIT study where the “95% AI projects fail” stat came from is flawed. Linking my post below.

In 2025, my team hosted a series of AI executive dinners.

We had hundreds of AI leaders.
From some of the biggest enterprises in the world.
Across every industry. Across functions.

We created these rooms intentionally. Not for keynotes or pitches.
For candid discussions on how AI transformation is actually happening.

𝗪𝗵𝗮𝘁 𝘄𝗲 𝗹𝗲𝗮𝗿𝗻𝗲𝗱: 
- When we curate "safe spaces" for peers, execs stop talking about the hype.
- They share what's working and what's not.
- They ask the real questions they can't ask anywhere else.
- They want to know what they don't know.

That's why we intentionally bring together diverse groups of enterprise AI executives and even include a handful of proven startup founders.

And these connections spark the real breakthroughs.

𝗟𝗼𝗼𝗸𝗶𝗻𝗴 𝗳𝗼𝗿𝘄𝗮𝗿𝗱 𝘁𝗼 𝗺𝗼𝗿𝗲 𝗼𝗳 𝘁𝗵𝗲𝘀𝗲 𝗰𝗼𝗻𝘃𝗲𝗿𝘀𝗮𝘁𝗶𝗼𝗻𝘀 𝗶𝗻 𝟮𝟬𝟮𝟲!

The NYC dinner tonight filled up within days.

𝗡𝗲𝘅𝘁 𝘀𝘁𝗼𝗽𝘀 𝘄𝗶𝗹𝗹 𝗯𝗲: 
➡️ London: Feb 23
➡️ Paris: March 10
DM me if you want an invite.

Which city should we host these in next?

I just learned my posts on LinkedIn have reached 500,000+ people and been viewed millions of times.

And honestly… that blows my mind.

I’ve always been a nerd who loves geeking out about AI.
It gives me energy.
It makes me curious.
But for the longest time, I never shared any of it online.

Starting was uncomfortable. 
And in those days, a few of you showed up with comments, questions, likes, DMs.
You have no idea how much that kept me going.

Because of this community, I’ve:
- Met incredible new people.
- Reconnected with people.
- Learned from your perspectives.

Whether we’ve been connected for years or recently: thank you for being here.
Thank you for engaging with me, challenging me, and teaching me.

We’re still in the early innings of AI adoption and real value creation.
Let’s keep learning, growing, building, and transforming together.

And for everyone in the US celebrating Thanksgiving:
In our fast-paced, ever-changing, AI-centric world, I hope you find a moment to pause, breathe, and be present with the people who matter most.

Happy Thanksgiving 🍁
Grateful for this community 🌟

Info on how to see your stats in comments.

🚀 If you’re building AI agents, Microsoft just changed your roadmap.

At Microsoft Ignite 2025, we announced new capabilities to help you build, scale, and govern agents in the enterprise.

Top 6 you need to know:

Agent 365
- The new control plane for the agent era.
- Unified view to register, govern, and secure every agent in your org.

Foundry Agent Service
- The production runtime to host, scale, and orchestrate your LangGraph, CrewAI, or custom agents. Comes with multi-agent support
- Built-in memory, monitoring, and autoscaling.

Entra Agent ID
- Verifiable identity + lifecycle management for agents.
- Helps solve the “shadow agents” problem and enforces zero-trust guardrails.

Unified context layer (Work IQ + Fabric IQ + Foundry IQ)
- Single semantic and knowledge foundation for retrieval, grounding, and agentic RAG across Microsoft 365, Fabric, AND Azure data.

Foundry Control Plane
- Observability + governance for every AI deployment.
- Brings policy, cost, risk, and performance all in one place.

Unified MCP Tool Catalog 
- Standardized tool discovery and connection.
- Helps connect agents to 1400+ enterprise systems (SAP, Salesforce, ServiceNow, etc.).

Together, these are building blocks for the enterprise AI agent stack.
Helping build secure, governable, and production-ready agents on day one!

🔗 More info on these announcements: https://lnkd.in/eqSSni7B
Additional agent-specific highlights in comments.

💬 What's your favorite announcement?

#MSIgnite2025 #MSFTadvocate Microsoft Developer

The big shift most AI teams miss:

Enterprises are no longer trying to build everything themselves.

A year ago, enterprise AI efforts were roughly split between build vs buy.
Today, 76% of AI use cases are purchased, not built in-house.

𝗪𝗵𝘆?
- Enterprises overestimated how fast they could build AI internally
- The 3P AI ecosystem is maturing fast
- Leaders are prioritizing speed to value over proprietary control

I see this daily working with the AI startups in the Microsoft for Startups Pegasus program.
They’re being adopted by some of the world’s largest organizations.

𝗧𝗮𝗸𝗲𝗮𝘄𝗮𝘆: 
- The winners won’t be the teams that build the best AI.
- They’ll be the teams that deploy value the fastest.
- Make sure you assess build vs. buy before moving forward.

💭 𝗜’𝗺 𝗰𝘂𝗿𝗶𝗼𝘂𝘀: What are you seeing companies do? Are they building, buying, or somewhere in between?

-
♻️ Repost to share with others that need to know this.
➕ Follow Heena Purohit for similar posts.

Source: Menlo Ventures.
🔗 More Enterprise AI trends and info in comments.

🚨 Big news: Anthropic Claude AI is now available on Azure!

Starting today, Claude Sonnet 4.5, Haiku 4.5, and Opus 4.1 models are available in Microsoft AI Foundry.

Azure is building the most open AI platform, giving AI builders:  
- More intelligence 
- More choice
- More compute

This is a HUGE unlock for teams building AI apps and agents! 
You have access to all frontier AI models on the same platform.  
That's secure, responsible, and enterprise-ready.

Full announcement: https://lnkd.in/eTrbKnQZ

PS: The Ignite keynote hasn't even started yet! Stay tuned for more 🙌

I posted this less than a week ago.

And it's already outdated:
- Kimi K2.5 released
- Anthropic launched Opus 4.6

Both are already topping production leaderboards.
And February will have many more anticipated model releases.

This is the new normal.

Technology is improving at unprecedented rates.
But it's also being adopted at unprecedented rates.

𝗪𝗵𝗮𝘁 𝘁𝗵𝗶𝘀 𝗺𝗲𝗮𝗻𝘀 𝗳𝗼𝗿 𝗔𝗜 𝘁𝗲𝗮𝗺𝘀:
Don't over-optimize on any one model.
Optimize for switching models.

Build a process that lets you test new model releases fast.
Measure what actually matters: not just data science metrics, but user and business metrics.

𝗧𝗵𝗲 𝗯𝗲𝘀𝘁 𝗔𝗜 𝘁𝗲𝗮𝗺𝘀 𝗮𝗿𝗲𝗻'𝘁 𝗽𝗶𝗰𝗸𝗶𝗻𝗴 𝘁𝗵𝗲 𝗿𝗶𝗴𝗵𝘁 𝗺𝗼𝗱𝗲𝗹.
They're the ones who can swap models in hours or days. Not weeks. Not a quarter.

📌 PS: This is exactly why I don't post "top AI models" or "top AI tools" lists. But let me know if you'd like to see more similar posts.
📌 My list of top anticipated model releases for February in comments.