Join the event trusted by corporate leaders for almost two decades. VB Transforma connects people building AI Real Enterprise. Learn more
Enterprises can be concerned about the influence of AI application when introducing into production, but the impediment to these projects with the lend a hand of a handrail at the beginning may snail-paced down innovation.
Andrew Ng, founder Deeplearning And And one of the most essential characters in the development of AI, emphasized the importance of observation and handrails in the development of AI during the Fireside chat VB Transform Today. He added, however, that they should not cost innovation and growth.
NG suggested that enterprises are building in the sandbox to quickly prototype projects, find pilots who work and begin to invest in observation and handrails in these applications after they prove that they work. This may seem contrary to intuition for enterprises that want to implement artificial intelligence.
>> See all our transform 2025 coverage HERE“There is an important role in observation, safety and handrails,” said NG. “To be honest, I tend to place them later, because I think that one of the ways in which large companies stop is that engineers are trying anything, they have to log out by five vice presidents.”
He added that vast companies “cannot afford that some random innovation team sent something that damages the brand or has confidential information”, but it may also hinder innovation.
Instead, NG said that sandboxes offer developers to teams to “only quickly with limited private information.” Sandboxes allow you to invest only in the projects that work and then add technology to make them responsible, including observation tools and handrails.
It often happens that enterprises establish novel sandboxes, especially for AI agents. Sandboxes allow for innovation within enterprises without touching any confidential information that they do not want to be public. However, they also allow bands to be as original as possible to test ideas.
Observation quickly becomes a key topic, because many AI applications and agents enter production. Salesforce He recently updated his agent library, AgentForce 3 to ensure better visibility in the agent’s performance and further support of interoperability standards, such as MCP.
Speed and lower costs of the pilot
In the case of NG, speed and innovation go hand in hand, and enterprises should not be afraid of it.
“Imagine that we were on a mountain queue, but this is a slow mountain queue. What happened in the last year, our mountain queue simply gathered high speed, and it is really exciting because it goes forward,” said NG. “I feel that the world is now on a very fast mountain queue and is great.”
NG said that one of the factors that contributed to this speed are tools available for programmers for quick work and ID, indicating that agents coding like Windsurfing AND GitHub Copilot I shortened the development time of “projects that took me three months and six engineers”.
These platforms of encoding agents and other tools that lend a hand programmers move faster, also meant the costs of performing pilot projects.
“I don’t think that the cost of proof of the concept is so low that I don’t mind making many poc (proof of the concept) is bad,” he said.
Barrier
However, one barrier can find talent. NG admitted that there are AI companies recruiting funding engineers of foundation models with a salary range of up to $ 10 million, but the price is not so high for software engineers.
“One of the biggest challenges for many companies is talent,” he said. “Good news for companies looking for engineers capable of building applications, the price is not near $ 5 million,” he said.
The problem, however, is that there is still not enough talents who have experience in building AI projects for enterprises. So NG returns to its first solution: let them experiment in a sandbox and gain this experience.