Most enterprises have a GenAI pilot. Far fewer have a GenAI result. Ahmad Fattahi, Executive Director of Generative AI at Genentech, spends his days closing that gap.
In this episode of The Builders Club Startup Founders Podcast, he explains what scaling generative AI really takes inside a large organisation, and why the biggest gains are still ahead.
How do enterprises scale generative AI beyond the hype? Ahmad Fattahi’s core view
Ahmad’s central point is that technology is rarely the bottleneck. In large enterprises, people, processes and the data they produce are what slow things down, so scaling generative AI is mostly an organisational challenge.
He also argues most companies are still in the early phase: attaching GenAI to existing workflows for modest, real gains. The explosive growth, he believes, comes when teams rethink the whole process around the new technology.
- Bottom line first. The earliest wins are efficiency gains: the same function done with far fewer person-hours.
- Top line comes later. Revenue growth and net-new products arrive later in the maturity curve, and can be roughly an order of magnitude bigger.
- People and process are the blockers. IT, business and data teams must pull in the same direction.
- Reinvent, don’t just retrofit. Replacing a tool inside an old process captures only part of the value.
What is the difference between AI and generative AI?
Ahmad sees AI as a continuum in which humans have gradually handed more decision power to machines, from paper maps to navigation apps to self-driving cars. The ChatGPT moment, in his view, was a genuine jump along that slope.
Earlier machine learning mostly classified, clustered or forecast. Generative AI produces what feels like new text, images or video from natural-language instructions, with little guidance.
- Adjustable creativity. You can tune how safe or inventive the output is.
- Context-aware. It can work within the context you give it.
- Semi-autonomous. It can reason, act and revise its own decision when something fails.
Where does generative AI create the most business value?
Ahmad slices impact into bottom line and top line. Bottom-line impact is the obvious first step: a team does the same work with a fraction of the effort, sometimes with better quality.
Top-line impact needs a bigger mental shift. It can help field teams find opportunities, stretch limited bandwidth, or support products that were not possible before.
The textile mill analogy
He borrows a historical parallel: early mills replaced water wheels with steam engines but stayed beside the river. The real boom came when owners moved factories closer to ports and population centres.
His point is that most companies today are still beside the river, adding GenAI to unchanged processes.
Which industries are adopting generative AI fastest?
Ahmad has seen life sciences, high-tech manufacturing, energy and financial services up close. He says life sciences has been surprisingly at the forefront despite heavy regulation, with chipmakers also moving quickly and energy more cautious.
He offers two reasons life sciences leads: a large share of its data is unstructured, such as medical documents and regulations, which suits GenAI, and competitive dynamics between providers and payers push both sides to adopt.
- Unstructured data. Specialised documents, regulatory text and clinical notes play to GenAI’s strengths.
- Provider and payer dynamics. Better, faster claims from providers pressured payers to match.
- Science itself. AI-driven protein research is raising hopes for faster drug discovery.
What actually stops enterprises from adopting AI?
In his experience, the blockers are rarely technical. Data, which is a byproduct of people and processes, is a common issue, as is getting IT and business functions working in unison.
He still says the train has left the station. He shares a second-hand example of a financial institution whose regulatory-form function reportedly dropped from about $600,000 a year to $10,000 with a GenAI vendor, while noting that failures such as the Air Canada chatbot case show the technology is imperfect.
- Map the people and process changes before choosing the model.
- Fix data readiness alongside the technology.
- Align IT and business owners early.
- Move early so you shape adoption rather than react to it.
Which AI tools does Ahmad recommend for professionals and builders?
He suggests mainstream assistants such as ChatGPT, Claude, Gemini and Microsoft Copilot, and says many people still underuse them for everyday decisions and planning. Enterprise users should check that a tool is sanctioned by their organisation.
For heavier learning and building he points to two options.
- NotebookLM. Load books, sites and videos into one notebook and get summaries, podcast-style audio or slide decks.
- CrewAI. Build agent teams by drag-and-drop or plain language, without being a programmer.
What does Ahmad think about AGI, singularity and the future?
He treats singularity headlines as good drama and is not losing sleep over it. He sees change as slopes with occasional jumps, and notes that AI systems are still far from being connected to each other or the physical world.
He does stress real needs for ethical AI and regulation, especially for military use or gene editing. What excites him most is the falling cost of intelligence, which could mean more personalised medicine and treatment for underserved areas.
Further reading and sources
- Genentech official website
- Generative artificial intelligence on Wikipedia
- Genentech on Wikipedia
- CrewAI official website
Frequently asked questions
How do I start scaling generative AI in my company?
Start with a high-volume function where person-hours are easy to measure and prove the efficiency gain. In parallel, work on data readiness and align IT and business owners, because Ahmad Fattahi sees people and process as the usual bottlenecks. Then look for chances to redesign the workflow itself.
Is generative AI worth the investment for enterprises?
Ahmad points to two kinds of return: bottom-line savings that show up first, and top-line growth or new products that come later and can be much larger. Results vary by case, so pilot against a clear metric before scaling.
Which AI tools should non-technical professionals try first?
Ahmad suggests a mainstream assistant like ChatGPT, Claude, Gemini or Copilot for daily tasks, NotebookLM for digesting large amounts of material, and CrewAI to experiment with agents. Check your organisation’s policy before using any tool with company data.
Conclusion
Ahmad Fattahi’s message is practical: the technology is ready, and the harder work is changing people, process and data. Retrofitting GenAI into old workflows pays off, but rethinking the workflow is where the larger gains sit.
If you lead a team, pick one process this quarter and ask what it would look like if designed around GenAI from scratch.
Want more conversations like this? Join The Builders Club, our community of 35,000+ founders and CXOs, and explore more episodes of the Startup Founders Podcast.
