OpenAI just had its demo a few days ago and everyone was blown away by the features of its latest upgrade. The coding is exponentially better, real time translation, application for a screen co-pilot and the shared experience of live video are all apart of this latest upgrade. Of all the amazing features that open AI released the number one thing multiple execs were most excited about was the cost. GPT-4o is 50% less than its predecessor. While conversations from the public was focused on the interpersonal features (which are amazing), the people that run the largest companies in Texas, US and the world were focused on the bottom line, 50% cost reduction in cloud calls.
Until local models become more efficient, the most popular way for enterprise to utilize AI is through the cloud. Here are some stats surrounding rising cloud costs according to Garter:
over 50% of SMEs tech budgets will go to cloud spend
Global end user spending on public clouds have increased from $421B, $500B and $599B over the last 3 years. Next year its projected to be $725B
Cloud costs are higher than expected for 60% of organizations to the 1,000 companies that responded
42% of CIOs and CTOs consider cloud waste the top challenge.
Over half of enterprises are struggling to see cloud ROI (PwC)
Yet in spite of all of that, money still keeps getting dumped into the cloud because they kind of have to. Compound that with the current market adoption of AI, the costs should continue to increase steadily if not exponentially. If executives choose to continue on this path of necessary spending in cloud then where can the cost savings come from? Building on what I wrote about previously wrote about, let’s examine human capital as an expenditure. I want to revisit this case because I think it’s more significant than previously thought. Organizations often start cost cutting measures by identifying underperformers and considering their dismissal. For the sake of not being hyperbolic let’s assume we are talk about 1 underperforming software engineer. If market adoption of AI continues, then that means the next phase of adoption will be AI agents. Lets dig into some numbers.
Costs
Human Software Engineer
According to the U.S. Bureau of Labor Statistics (BLS) and other sources, the average annual salary for a computer engineer in the United States is approximately $117,000 (as of 2023).
Let’s assume a standard full-time work year is about 260 days (52 weeks x 5 days/week, excluding holidays and vacation).
$450 daily cost for an engineer
AI Software Engineer
Total Daily Operational Costs (Estimated) = $280
Cloud Computing = $100
Storage = $20
Data Transfer = $10
Monitoring and Support = $80
Licensing Fees = $50
Energy Costs = $20
365 days a year since the computer could run 24/7. For this exercise we will keep it to 260 days.
Productivity
Human Software Engineer
4-6 hours of productivity from standard 8 hour day (Let’s do best case scenario 6 hours)
AI Software Engineer
AI Agent can effectively operate at or better than a human programer for 24 hours a day.
An AI agent could perform these tasks up to 10 times faster than a human, assuming no need for breaks and minimal errors.
Assuming the AI is advanced, it could match or exceed human productivity, performing tasks 1.5 to 2 times faster on average due to the lack of context-switching and fatigue.
In order to quantify the output we will make (6 hours = 1 Unit)
Human Engineer
1 unit of productivity a day
260 per year
$450 per unit
AI Engineer
4 units of productivity a day
1040 Units Per Year
$70 per day
Individual Cost
Weighted cost analysis of Human engineer vs an AI engineer over 5 years
Now over the span of 1 years lets compare the cost for a human engineer to match the output of a AI agent from a unit perspective.
Team Cost
Now lets scale this out to just one full engineering team.
Team 1Human Engineers:
Annual salary for each engineer: $117,000
Total annual cost for 6 engineers: 6 * $117,000 = $702,000
Units produced per engineer per year: 260 units
Total units produced by 6 engineers per year: 6 * 260 = 1,560 units
Team Lead:
Assuming the team lead earns 20% more than a standard engineer: $117,000 * 1.2 = $140,400
The team lead production is not considered in unit calculations due to multiple obligations in the company and the focus on building frameworks.
Total Annual Cost for Human Team:
Engineers: $702,000
Team Lead: $140,400
Total: $702,000 + $140,400 = $842,400
Team 2
AI Agents:
Annual operational cost for each AI agent: $72,800
Total annual cost for 6 AI agents: 6 * $72,800 = $436,800
Units produced per AI agent per year: 1,040 units
Total units produced by 6 AI agents per year: 6 * 1,040 = 6,240 units
Team Lead:
Same as above: $140,400
The team lead's production is not considered in unit calculations.
Total Annual Cost for AI Team:
AI Agents: $436,800
Team Lead: $140,400
Total: $436,800 + $140,400 = $577,200
Annual Cost Savings
Team 1:
Total units produced: 1,560 units
Cost per unit: $842,400 / 1,560 ≈ $540 per unit
Team 2:
Total units produced: 6,240 units
Cost per unit: $577,200 / 6,240 ≈ $92.52 per unit
Annual Savings on a Per Unit Basis
Team 1:
Cost to produce 6,240 units: 6,240 * $540 = $3,369,600
Team 2:
Cost to produce 6,240 units: $577,200
Annual Cost Savings:
Savings: $3,369,600 - $577,200 = $2,792,400
Cumulative Costs and Units Over 1 to 5 Years for 1 Team
Keep in mind that these numbers would be even higher considering the AI could work 24/7, doesn’t need sick days, doesnt take vacation ect… Here are what the numbers realistically look like. I have taken out 14 days for maintenance, upgrades and other factors.
Conclusion
When one digs into the numbers even more, the unrealized financial gain of this technology should leave the C suite no choice but to pursue this type of transformative technology. The return on value is more lucrative the more you scale. Even though I am a proponent of this tech, I am not a proponent of fire everyone and only run agents. I believe the best way to utilize this is to empower your high performers to achieve a higher productivity output. The companies that are first adopters of this technology will gain not only a significant production output advantage but also a significant competitive market advantage. This is where the market is going and as my friend likes to remind me, technology is undefeated. When it comes to building these agents, fingers crossed we are one of the first ones.
