95% of enterprise ai agent projects die in pilot – and it's not a model problem

95% of enterprise ai agent projects die in pilot – and it's not a model problem

manoj saxena, who took ibm watson to market and now builds ai governance at trustwise, on the @craigss eye on ai podcast:

building an agent takes hours, getting it past compliance, security, and finance takes 6-7 months. one input into an agent can trigger 20-50 actions and consume 20-40x more tokens than gen ai two years ago – for some enterprise agents, the annual operating cost already matches a full-time employee.

per cloudflare, 57.4% of all internet traffic is now bots. saxena compares it to when data traffic overtook voice on at&t – and says the internet will never be the same. he expects 90% of users of their ai control tower will be agents, not humans, within three years.

the cyber trust market: $2b now, $80b by 2030. the next enterprise ai problem isn't building agents – it's controlling them while they operate.

full segment via link. source: eye on ai podcast