apidays Helsinki & North 2025 - Agentic AI: A Friend or Foe?, Merja Kajava (Aavista)

Slide 1

Slide 1 text

Aavista Oy Merja Kajava apidays Helsinki & North 2025 Agentic AI – A Friend or a Foe

Slide 2

Slide 2 text

“What is the carbon footprint of a ChatGPT query compared to a Google query?” Photo credit: NASA / Alex Gerst

Slide 3

Slide 3 text

Vanderbauwhede (2024): “AI-generated answers to conventional search queries dramatically increase the energy consumption. By our estimates, energy demand increase by 60-70 times.” Vanderbauwhede, W. (2024). Estimating the Increase in Emissions caused by AI-augmented Search. https://doi.org/10.48550/ARXIV.2407.16894

Slide 4

Slide 4 text

Moving from chatbots to Agentic AI

Slide 5

Slide 5 text

Tools Orchestration Model Integrate to storage and 3rd party services Reason with goals Goals, profiles and guidance What are the elements of Agentic AI

Slide 6

Slide 6 text

Connecting agents

Slide 7

Slide 7 text

Agent-to-Agent (A2A) by Google Model Context Protocol (MCP) by Anthropic “UDDI for Agents” for agent discovery “BPEL for Agents” for orchestration NLWeb by Microsoft Agent Agent Agent Tool Tool Tool RSS Schema.org Protocols are emerging for agent connectivity MCP MCP MCP A2A A2A MCP API MCP Servers MCP Servers

Slide 8

Slide 8 text

Sustainability AI governance Security and privacy Testing Performance Data quality Cost

Slide 9

Slide 9 text

Nothing exists until it is measured Niels Bohr

Slide 10

Slide 10 text

Carbon footprint Cost ~

Slide 11

Slide 11 text

Calculate the carbon footprint Cloud providers Data centers Service providers IPaaS API SaaS AI AWS Google Cloud Microsoft Multi-cloud Applications Carbon footprint calculators by cloud providers. Multi-cloud calculators. Open-source calculators, for example Code Carbon. Commercial calculators, for example from Dynatrace and Splunk.

Slide 12

Slide 12 text

“Processing a single query might involve over 50 LLM API calls. The calls tend to be very narrow and specific. Different kinds of calls may be to different models.” NLWeb – Chat query example https://github.com/microsoft/NLWeb/blob/main/docs/life-of-a-chat-query.md Case NLWeb – Chat query example

Slide 13

Slide 13 text

Choose models carefully

Slide 14

Slide 14 text

AI Energy Score - Extractive Q & A by Huggingface Each Gen AI model is different Luccione et al. (2024): “Multi-purpose models are more energy-intensive.” Luccioni, S., Jernite, Y., & Strubell, E. (2024). Power Hungry Processing: Watts Driving the Cost of AI Deployment? The 2024 ACM Conference on Fairness, Accountability, and Transparency, 85–99. https://doi.org/10.1145/3630106.3658542

Slide 15

Slide 15 text

Multimodal AI brings new challenges Multimodal streaming “Agent architecture with bi-directional event streaming running 24/7”

Slide 16

Slide 16 text

Multimodal AI From text prompt to images,videos and speech Luccioni, S., Jernite, Y., & Strubell, E. (2024). Power Hungry Processing: Watts Driving the Cost of AI Deployment? The 2024 ACM Conference on Fairness, Accountability, and Transparency, 85–99. https://doi.org/10.1145/3630106.3658542 “Tasks involving images are more energy- and carbon-intensive compared to those involving text alone.”

Slide 17

Slide 17 text

Energy certificate for AI agents Photo credit: Energiakartoitus Oy

Slide 18

Slide 18 text

Key takeaways Choose the minimal solution YAGNI Follow up the costs Measure the carbon footprint