The Founder`s Code podkast

AI Is Getting Smarter. Why Do AI Projects Still Fail? | Vishakha Gupta, ApertureData

0:00
52:51
Do tyłu o 15 sekund
Do przodu o 15 sekund

Some AI teams have three months to prove a project works. Vishakha Gupta has seen the data setup alone take six to twelve. When the pilot disappoints, the conclusion is often: “AI doesn’t work.”


On The AI Floor, Shamil Malachiyev sits down with Vishakha, founder and CEO of ApertureData, to look at what actually stands between an impressive AI demo and a system a company can use. After seven years at Intel Labs and years building infrastructure for multimodal data, she has seen the same problem from research through to production: a better model cannot help much if it cannot access the right data, in the right form, at the right time.


The conversation starts with siloed product images, catalogs, inventory, and customer information. It moves through the cost of AI infrastructure, the gap between a proof of concept and a secure deployment, and the next difficult question: how do you give AI agents useful memory and context across an entire company?


Expect to learn:

- Why assembling the data for an AI project can take 6–12 months

- Why a pilot can work on sample data and still fail in production

- What teams should settle before choosing models and tools

- How the cost of a data system changes from a small deployment to enterprise scale

- Why junior AI talent and experienced system architects need each other

- Why knowing what happens underneath an AI tool still matters when agents can write code

- Where today’s models still struggle with images, video, and complex real-world data

- How Vishakha separates genuine AI safety concerns from attention-grabbing headlines

- The difference between knowledge, memory, and cognition in an AI agent

- Why an agent needs to know who provided information, when it was true, and who is allowed to see it


If you’re building AI inside a company, this is a conversation about the work that begins after the demo.


CHAPTERS


0:00 Intro: Vishakha Gupta and the challenge beneath AI

1:11 From Intel Labs to building ApertureData

5:37 AI meets the wall of real-world company data

6:16 Why useful data lives in different departments

10:36 When a weak pilot becomes “AI doesn’t work”

13:10 What successful AI pilots do differently

17:07 What AI data infrastructure actually costs

20:49 AI jobs, new graduates, and experienced architects

24:43 Do you still need to understand what’s under the hood?

29:43 Why a proof of concept is not a production system

30:07 Are models moving faster than AI adoption?

31:17 What today’s models still struggle to understand

38:30 AI safety warnings: real risk or marketing?

42:55 What memory and cognition mean for AI agents

47:35 Explaining AI cognition through a novel

51:07 Where to start if you want to learn AI

53:20 Vishakha’s “council of agents”


👇 CONNECT WITH VISHAKHA GUPTA

Vishakha Gupta ► https://www.linkedin.com/in/vishakha/

ApertureData ► https://www.aperturedata.io/


👇 CONNECT WITH SHAMIL

Shamil Malachiyev ► https://www.linkedin.com/in/shamilmalachiyev/

FluidLabs ► https://fluidlabs.com/


🎙 LISTEN to The AI Floor (The Founder's Code, Season 2)

Spotify ► https://open.spotify.com/show/7v6YOQ77R1xzbOiiHsDawd

Apple ► https://podcasts.apple.com/sa/podcast/the-founder-s-code/id1797768317


🔔 SUBSCRIBE for conversations with the people building and deploying AI in real companies.


ABOUT THIS CHANNEL

The AI Floor brings you ground-level conversations about what is actually happening with AI and automation across the industry: the breakthroughs, the difficult decisions, and the work behind the demos.


#AI #EnterpriseAI #AIAgents #AIMemory #ApertureData #TheAIFloor

Więcej odcinków z kanału "The Founder`s Code"