Become a Creator today!Start creating today - Share your story with the world!
Start for free
00:00:00
00:00:01
May Habib: Why Agentic AI Needs Guts, Grit - and the Human Touch
175 Plays16 days ago
Rewrite the Rules or Be Left Behind
with May Habib, CEO & Co-founder of Writer
“This is GenAI’s COVID moment—don’t waste the crisis.” — May Habib
What You’ll Hear in 30 Minutes
• Why enterprises are still miles away from building real agentic systems
• How Writer’s Palmyra X5 enables self-evolving AI agents with memory
• What smaller firms can teach enterprise giants about speed and scrappiness
• The new formula for promotions in the age of automation
• Why “loving tech and loving people” is the ultimate success metric
Guest Snapshot
May Habib is the co-founder and CEO of Writer, an enterprise-grade LLM platform focused on AI agents that work. From translation tech to leading-edge self-evolving models, Writer’s Palmyra models power agentic AI for the world’s top brands—from Salesforce to Spotify. Habib shares how her early work in machine translation shaped Writer’s agent-first architecture and why the next leap in generative AI isn’t technical—it’s cultural.
Timestamps:
00:00 – What Sets Writer Apart
02:35 – From NLP Roots to Real-World Agents
04:35 – Why Now: AI Urgency & Enterprise Inertia
07:45 – Enterprise Adoption: Challenges & Gaps
10:31 – Agentic AI in Action (Mars, Prudential, Salesforce)
14:27 – The Rise of Digital Personas & AI-Augmented Work
18:59 – Redefining Leadership & Recognition in the AI Era
22:00 – Scaling Up: Funding, Talent & What’s Ahead
27:16 – Culture, Grit & Staying Close to the Customer
Learn more about Writer: https://writer.com
Follow May Habib: https://www.linkedin.com/in/mayhabib
Follow Phil Fersht: https://www.linkedin.com/in/pfersht
Loved the conversation? Follow, rate ★★★★★ & review From the Horse’s Mouth to keep bold, practical AI insights flowing to your feed.
Recommended
Transcript
Introduction to 'From the Horse's Mouth'
00:00:12
Speaker
You're listening to From the Horse's Mouth, intrepid conversations with Phil First. Ready to meet the disruptors who are guiding us to the new great utopia by reshaping our world and pushing past corporate spin for honest conversations about the future impact of current and emerging technologies?
00:00:30
Speaker
Tune in now.
Meet May Habib, CEO of Writer
00:00:37
Speaker
Welcome to the latest edition of The Horse's Mouth. My name's Phil First. on the I'm the host of the podcast. I'm also the CEO at HFS Research. And i'm today I'm delighted to have May Habib join the conversation.
00:00:51
Speaker
I'll let you introduce yourself, May. Can you tell us a bit about the role you're doing with Writer, but a little bit about the vision that drove the inflection of the firm and and that sort of thing?
The Journey and Innovation of Writer
00:01:01
Speaker
Yeah, happy to, Phil.
00:01:03
Speaker
Such a fan of you and HFS Research for a while and really excited to be with you and your audience today. Writer is an agent builder platform for the enterprise. And i think the last four years for us, since we started the company, of five years now, has really been about productizing as furiously as the technology evolves.
00:01:27
Speaker
We build our own large language models, our family of Palmyra models are state-of-the-art. And for every big step function change in what the models can do, we have been there to really productize that offering to to the enterprise. And our last and latest model, Palmyra X5, is a rag-killing model.
00:01:50
Speaker
agent enabling machine. And i think so much of what we've all been just talking about for the past year or so when it comes to agentic is now absolutely possible.
Generative AI: A Transformational Point
00:02:01
Speaker
We work with the industry's most innovative AI teams, folks, Uber, at Franklin Templeton, ah Vanguard, ah Salesforce. And think we probably have more agents in production than literally any company or hyperscaler.
00:02:18
Speaker
in the industry. The result of five years of architecture investments that have made it possible to build these big AI orchestrations representing some of the most important workflows in our customers' businesses. So a whirlwind of a five-year journey, Phil. I feel like we've been four or five companies in that timeframe.
00:02:42
Speaker
That's incredible, isn't it? i mean, so when you... worked, I think, with Cordoba. How did that influence the direction that you've taken writer in? Yeah, well, my first company with Wasim, my brilliant co-founder, was in the machine translation space. And folks acts ask me all the time, how do you build LLMs now? Like, what was the crazy path you were on that led you here?
00:03:04
Speaker
ah Wasim and I have been in NLP for, you know, a decade and a half now. And it really was the NLP journey. Across AI engineering and LLM development, you find all sorts of the previous generation of NLP researchers and engineers, because it really is how and where these attention algorithms really got got their start. And in the machine translation space, you know, we went from using statistical approaches to deep learning approaches to encoder-decoder approaches to the translation problem. And it's really when we we stripped out the decoder technology and really said, look, this is a very different kind of business is when we started Writer in 2020.
00:03:47
Speaker
and twenty twenty Now Palmyra leads on all sorts of blue scores, etc., BLEU, when it comes to kind of quality of the translation that actually happens through the LLM.
00:03:59
Speaker
And we've got translation use cases all over our business. So it's fun to kind of be in like the translation game still, but with such powerful technology that really is software that builds software, right? And this is why the company is called Writer. If you can write it, you can build it.
00:04:16
Speaker
And you can basically write yourself a localization platform now using LLMs. So it's been fun and really has felt like one story, you know, from the beginning of our entrepreneurial life.
Enterprises and the AI Revolution
00:04:28
Speaker
Well, I mean, this really does follow on from our summit where last week where we really feel that what was a smoldering platform last year has now become a burning platform. I think the impact of tariffs, geopolitical uncertainty is actually driving enterprise leaders to look at how do we do more virtually? How do we maybe reduce our reliance on these markets?
00:04:53
Speaker
huge global supply chains and and become more independent and in control of our data. What are you seeing in your own conversations, the fact that you're leading the way with translation in particular and these emerging agentic models?
00:05:08
Speaker
What are you seeing from clients? Are you seeing increased urgency at the moment? I think this is going to be generative AI's COVID moment. You can never waste a crisis in the enterprise. that The amount of concrete wall-busting guts and confidence you got to have to completely really rewire big workflows to be agentic first, AI first,
00:05:31
Speaker
So few people are doing it, Phil. Like, there's so much talk about interoperability and future-proofing, and the vast majority of the IT organizations out there haven't built an AI agent that can string two API calls together. So, you know, we are absolutely, I think, in this very interesting confluence of Folks understand that they've got to lead their companies, right? CIOs are really understanding it's on them to lead their companies into this future.
00:06:03
Speaker
And at the same time, I think the chasm between where they want to go and, you know, their ability technically and from a tooling perspective to get there. It's large.
Opportunities in Global Challenges
00:06:12
Speaker
And for our kind of super avant-garde, you know, on the bleeding edge customers, your Mars, your Kenviews, your Accentures, your Salesforce's, I do think they are looking at the geopolitical and economic landscape as an opportunity to accelerate what they're doing in generative AI internally.
00:06:33
Speaker
And I said for a while, you know, privately to folks that we need our COVID moment. You think about what happened to Zoom and collaboration tools and obviously a horrible global crisis that we all wish didn't happen.
00:06:46
Speaker
And it was a boon to the kind of tooling and work that we now could not see our lives without. And it led to a measurable degree of flexibility and efficiency, being able to support hybrid workforces.
00:07:00
Speaker
I think moving to a world where we really do live side by side with agentic AI doing very meaningful work is going to require similar levels of external forces because it is just so much inertia that is standing between people and true agentic AI transformation, even once you get the technology
AI Adoption in Small vs. Large Enterprises
00:07:22
Speaker
there. I mean, we have such powerful agents in production and both us and our champions leading the charge internally wish they could do more, go faster, go deeper.
00:07:33
Speaker
But there is a huge element of needing to bring folks along. There is a huge element of just change management and program design Right. You build agents that can automatically update your website. Amazing.
00:07:46
Speaker
All right. Now we've got to get legal and compliance on board, too. Right. And it is a much larger set of conversations that I think need to be supported by the C-suite, need to be supported by the board in order to really go at the pace that the technology deserves.
00:08:02
Speaker
You talk about inertia, are you seeing it in large enterprise? And is this an opportunity for maybe smaller businesses, which can now punch above their weight to attack them? Or are you seeing it like right across the board?
00:08:14
Speaker
So the larger enterprises are definitely investing more, experimenting more. They've got larger sets of resources against this, no no doubt. But the mid-market can go much faster. We think about a sales process with a research firm, right, or a hedge fund.
00:08:34
Speaker
The smaller they are, and more likely they are to... share data and examples and business logic very quickly in a process that allows us to build really powerful agents as a demonstration, right? And those are just, you know, very technical examples, but certainly smaller companies have more at stake in a lot of ways that I don't think all of them recognize.
00:08:55
Speaker
The companies that are investing the most, learning the most, are going to eat up their rivals, big or small, And I think there's just a tremendous opportunity for increased market leverage, you know, the more you're able to deploy these agents inside of your businesses. So I'm surprised at how slow the mid-market seems to be, although we've got incredible customers in the private equity space who are realizing this and are working with us to build tooling for their portfolio companies
00:09:26
Speaker
as a way to really accelerate their development. So the PE shops who own a lot of the mid-market are a little quicker on the ball when it comes to Agentic. And, you know, I think their portfolios are really benefiting from that.
00:09:39
Speaker
Right. So, you know, you mentioned clients you got like Mars and Kenview, Intuit, these types of businesses, even
AI's Impact on Marketing and Sales
00:09:46
Speaker
Accenture. Can you share a little bit about how these organizations have integrated these AI solutions and what real benefits they're experiencing?
00:09:54
Speaker
Yeah, absolutely. So we've got a agent builder and orchestrator in Rider that allows for a hybrid of deterministic and non-deterministic technologies to come together.
00:10:08
Speaker
So if you are building an agent to do fund reporting and you are Franklin Templeton, for example, or you're Vanguard, You are orchestrating across multiple sources of structured and unstructured data, and you're now putting into an automated workflow the people and the process that needs to be followed before you've got, you know, the kind of fun report that can automatically go to a big institutional organization.
00:10:35
Speaker
client, for example. And the the kind of like swivel chairing that needed to happen between and across systems, the like just gross number of people that needed to be included in processes like this can really be incredibly and very efficiently and to many instances, higher degree of quality and personalization if you're able to agentify. And the ability for us to do this in a way that is high quality, where AI is making the right decisions, where this is not failing at scale, is quite differentiated.
00:11:12
Speaker
On the sales and marketing space, we've got customers, especially from a technology perspective, these guys can go so fast on digital journeys. Our customers include, on the tech side, Spotify and Uber and Intuit, HubSpot, NVIDIA, Salesforce,
00:11:28
Speaker
And really being able to reinvent marketing and sales processes, sales enablement processes, AI first, really agent first is really great. We use a lot of these kinds of agents internally as well, being able to automatically review and respond to RFPs, being able to automatically and agentically look at what salespeople are saying on calls, how that works.
00:11:52
Speaker
measures up against scorecards, what enablement material we should send them based on the gaps. It just really allows a reinvention of how we thought about these processes independent of the tools and the systems. So many sales and marketing leaders technology leaders have to think in terms of the confines of the systems they use and are no longer confined to that.
AI Agents as Personal Assistants
00:12:16
Speaker
Credential's go-to-market team, incredibly innovative. They've automated how they create campaign briefs. They've automated how they take those briefs and actually produce the assets dictated in the campaign briefs.
00:12:30
Speaker
And then they've automated how all of those pieces of collateral actually make it into their email system, their CMS, their personalization tools, their communication platforms, the impact of the kind of marketing you can do, the speed of that marketing, innovation of the campaigns, the personalization of the messaging, when you can actually kind of go up to the velocity, right, and the valence of what do I want to be doing with my marketing versus what's my email tool capable of.
00:13:03
Speaker
So in terms of where this is going, I've had some far-reaching conversations on you start to create agents that really become these wonderful assistants in helping you do your jobs. I mean, I'm an analyst. i I'm training LLMs all the time to help me do my job. And it it really is, it's mind-blowing how quickly this is evolving.
00:13:24
Speaker
But we're getting to the point now where Can you see people almost starting to create an agentic persona for themselves within organizations? Say, for example, you go on vacation, in those two weeks you're away, you have imprints of your network, your calendar, how you interact with people, work that you're performing.
00:13:46
Speaker
So people can actually interact with your own agent, even though you might not be talking them directly. talking with them directly Do you think it's going to go down that path eventually, or do you think agency is going to go in a different direction?
00:13:58
Speaker
I do think that our ability to be always on is going to take on a ah different meaning. I absolutely see the way that we work today is grossly underleveraged and painfully analog compared to what it is that that we can be doing.
00:14:17
Speaker
Anytime we're building an agent and we work full on really mission-critical types of workflows, what we're essentially doing is taking some knowledge of an intelligence of a company's workforce on a particular process and getting it into the hands of an agent or an agentic system.
Trust and Adoption Challenges in AI
00:14:39
Speaker
You are a manufacturing plant and you get audited on a regular basis. Your QA processes and your review readiness and QA readiness processes aren't written down for the vast majority of the organizations we work with. They might be, but they're out of date now or don't fully capture all the nuances, etc.,
00:15:03
Speaker
When you are building and testing these agentic systems to to do these processes, you are essentially filling in that gap, right? And getting it into an agent that can take over that work and really augment the humans who are doing that that work. And I think we're going to see that consumerized, right, down to replying to your inbox email while you're away on on PTO. So again, the courage, though, that it takes, right, to have that kind of faith and trust in technology, right?
00:15:32
Speaker
I don't know that we've gotten enough people, right, over that initial hump, right? The vast majority of folks using Copilot or ChatGPT are like, okay, this is cool, this is good, but replacing me or augmenting me or like speaking on behalf of me, I don't know if I trust, right, being able to to get there. So I do think it requires, you know, an AI system beyond what is happening in some of these chatbots for folks to really see the power and that the technology is actually there now.
00:16:00
Speaker
Yeah, and I think it can also increase someone's value and visibility to their organization in general or to the outside world because you're both you're more authentic. I mean, I write for a trade and I can have help with my writing using ChatGPT, for example, but ultimately,
00:16:19
Speaker
You're the soul, right? I know if something's been generated by a human being or by a computer. But where's that line? i feel like we're getting to a point where there's going to be a sort of line between what is private to an individual and what should be public to their organization, right? So is it their relationships? Is it...
00:16:38
Speaker
the IP they create. Because in a weird way, by agendifying a lot of the work you do, you're creating a lot more visibility and knowledge of what your own people are doing. It's a tremendous and you know and introverted approach to understanding your own business.
00:16:54
Speaker
But where is that line? And are we crossing it yet? Is it something that you're hearing raised a lot by your clients? We've got a lot of customers who are building proxies for their executives when it comes to amplifying their voice in market.
00:17:11
Speaker
So CEO of NVIDIA, CEO of Qualcomm, right? We've gotten a number of these in production and No one reads one of these LinkedIn posts or, you know, letters to shareholders, et cetera, et cetera, and goes, oh, you know, so-and-so most definitely did not write that, right? It's much more confident comms teams, right, that are now putting better first drafts, right, or second drafts in front of their execs before that's published,
Redefining Success with AI Contributions
00:17:39
Speaker
right? So much of the queasiness factor, I think, goes away when you don't just have a human in the loop. I don't love that phrase. You have a human at the helm.
00:17:48
Speaker
yeah but You know, the the questions to ask how to deploy this technology, it's got to be in the hands of people who know what excellent looks like because our jobs as humans is to constantly push that envelope, right? And so you said something, though, Phil, that I really want to come back to.
00:18:06
Speaker
How do we elevate and promote people and shine a light on their work? So much of the value of somebody inside of an enterprise today is the human headcount under them, is the agency budget they control, is the contractor budget that they've got.
00:18:25
Speaker
And I think the organizations we have seen be really successful at getting like both top-down and bottoms-up buy-in on agentic AI and and generative AI in general have been ones that have really tried to elevate people who think leverage first.
00:18:43
Speaker
impact first when it comes to their work. And I think before we really see the the widespread use of agents in the enterprise, we've got to really change the way we think about how we promote people and you know who we elevate, who we shine a light on.
00:19:00
Speaker
We try to really set the example of anybody who's built on writer agents that save the company money or you know produce much more impactful digital results or whatever it is There's celebration. There's Lucite trophies. There is recognition from the C-suite, right, of their accomplishments. There is elevation in stature and in title.
00:19:23
Speaker
And it has nothing to do with, oh, I now have more headcount. Our traditional ways of rewarding people are in direct conflict with encouraging them to use AI in innovative ways.
00:19:36
Speaker
I couldn't agree more, and I think it's very eloquently put on how we take this forward as human beings and workers.
Writer's Future and Global Expansion
00:19:43
Speaker
So looking ahead to the future, I think you guys last November and announced a big $200 million dollars funding round. That's a good chun ah chunk of change.
00:19:53
Speaker
Takes you into a multibillion-dollar valuation, I'm assuming. But how do you plan to invest that capital, What are you going to change or evolve with your products so that your clients can anticipate in the future?
00:20:06
Speaker
Yeah, absolutely. A big part of where we're investing that capital is in our approach to models with memory. Self-evolving models are our large language models with a really fundamental difference.
00:20:23
Speaker
The training data is not static and can continue to evolve once you deploy the model, hence the name. And that's really important for agentic AI, Phil, because X5 might have a massive multi-million token context window that's lossless, forgets nothing.
00:20:40
Speaker
It is best in class at tool calling and agent building. And... Right. Your ability to learn the more you use it is absolutely stunted if you can't update the model in real time.
00:20:53
Speaker
And our ability to say, all right, I've got a hiring agent that this is, you know, a use case that we've got internally, a hiring agent that automatically produces offers. Right. And DocuSign.
00:21:04
Speaker
And it made a mistake around the amount of equity to put in an offer because we've got a different table for engineers than for go to market than for AI researchers. Right. When a human lets the agent know that it made a mistake, right, how do we make sure that that knowledge is retained, right, and the next person that uses it is going to not get the same error?
00:21:27
Speaker
You're going to run out of that context window very quickly, especially, right, in knowledge-intensive use cases. And so for us, so much of the research effort on self-evolving models was really driven by our actual lived experience with agents and understanding realizing that we needed to take transformers in a new direction.
00:21:50
Speaker
So we're very excited about that. Those models, fingers crossed, will GA a this year. and of course, there's been a big go-to-market build-out of a writer, a huge demand internationally for what we do, agents that actually work.
00:22:04
Speaker
So we now have offices in London and Dublin and Singapore, hiring very aggressively in the U.S., opened offices in Austin and Chicago, as well as our New York and SF flagships.
00:22:15
Speaker
that was actually my next question, which was, where are you getting the talent? I mean, where are you finding the tech talent and the, i don't know, but the evolving thinkers to really help you get to that next level?
00:22:27
Speaker
Yeah, you know, in average, in venture-backed companies, folks are interviewing 15 to 20 people for every hire. We're interviewing 85 to 90 people for every hire.
00:22:39
Speaker
So, you know, we've got a very specific thing that we are looking for.
Leadership and Passion in Hiring
00:22:44
Speaker
It's a very high-density place, Ryder. And I think what we're looking for boils down to humans who love technology and love people just as much. This is a very innovation-first, human-first effort.
00:22:59
Speaker
We seek to understand um before we give advice, and especially in our customer-facing roles, it is very important that people are experts on the technology and know who we've been, who we are, and who we're becoming, right?
00:23:17
Speaker
And are able to actually explain that to the customer that's getting 40 pitches a week and be able to advise at the same time as being able to do that, advise on how and where they start with generative AI inside of their own enterprises.
00:23:31
Speaker
Those are their unicorns across every dimension. On the engineering front, we're hiring people who want to be on the phone with a customer you know for a part of the product that is zero to one.
00:23:42
Speaker
um In AI and in research, we're hiring people who really want to be close to the customer problem that we're solving. We get a lot of folks interviewing and applying who are kind of refugees from the big shops where it's felt like They were really divorced from, you know, the end user experience and the products that were built on top of their work.
00:24:04
Speaker
And so, you know, have built a team kind of across the board of very customer centric people. I guess then that's your advice to aspiring entrepreneurs, right? They've got to be customer-centric. And I love what you say about you have to love humans as much as you love the technology.
00:24:21
Speaker
That, to me, sums up where we are in this space and where we need to go. Just one final question, May. You said you've gone through five iterations of the firm since you've taken it over.
Reflections and Future Outlook
00:24:32
Speaker
What personal experiences have shaped your approach to the company culture that you're talking about so passionately? I think... Being close to the action and encouraging your leadership to be close to the action, to be close to what solid central truth is, it's just always been, you know, a factor and in how I lead and and how I approach my work life.
00:24:57
Speaker
The Space moves so fast that nothing here can be theoretical and our ability to lead the customer. I mean, i I see a big part of what the writer platform does, right? The full stack generative AI platform as abstract the rate of change for the market.
00:25:18
Speaker
And, you know, our customers repeatedly tell me they heard about graph-based drag from us before they read about it. They heard about synthetic data from us before they read about it. They heard about continuous training of models from us before they read about it.
00:25:32
Speaker
And we want to be the company that brings, right, the latest and greatest to these folks in a way that actually works, right? The vast majority of what's happening in the market right now is marketing, right?
00:25:45
Speaker
And folks are talking about things that they haven't written a line of code for yet. And we are the exact opposite, right? We are bringing things that actually work and together with the customer talking about how it would work for them in a way that just really resonates and is just very, very real.
00:26:02
Speaker
I don't know what would have shaped that way of working. i've I've always worked that way. I think, you know, I grew up in family businesses and just always saw my parents, my uncles, just very hands-on. And so maybe there was just there were are no professionals in my family. There were no professional managers. you know There were no doctors, lawyers, accountants.
00:26:22
Speaker
It was just blue-collar people were very hands-on. And I guess I've approached knowledge work that way too. I love that. I love that. My entire family are academics.
00:26:34
Speaker
So i'm i'm like I'm like the dark sheep. I'm the the one... I'm the bad apple. What did I end up doing? I'm running the business. Almay, this has just been wonderful. I think we'll have you back on again soon as well because there's a lot more to talk about. But I'm a really big admirer of how quickly you've propelled your company into the spotlight. You talked about a lot. You're a big example of identification and the future of the industry. So all the very best in the next few months as we go through this wonderful world of uncertainty that we live in. And hopefully we will come out the other side a little bit sharper, a little bit smarter and and a bit happier. So thank you very much for your time.
00:27:12
Speaker
Thank you, Phil. Thanks so much.
00:27:18
Speaker
Thanks for tuning in to From the Horse's Mouth, intrepid conversations with Phil First. Remember to follow Phil on LinkedIn and subscribe and like on YouTube, Apple Podcasts, Spotify, or your favorite platform for no-nonsense takes on the intricate dance between technology, business, and ideological systems.
00:27:37
Speaker
Got something to add to the discussion? Let's have it. Drop us a line at fromthehorsesmouth at hfsresearch.com or connect with Phil on LinkedIn.