![04.02.2025 15
Traits of an AI Agent
1. Communication capability
• Natural language
• Other technical systems
2. Skills / tools
• Specialized skills for the domain
• Mathematical functions, web-search, etc.
3. Independent operation
• Make decisions on the steps to take
• Decide on tool use
• Ask for clarifications
• Create larger plans, collaborate
• Like humans, operations are not without any oversight
4. Guardrails
• Sandbox for running code
• Read-only access to databases
• Error checking
Charles Watson
Financial Analysis
To: Finance Department
Subject: Request for Historical Financial Data (2021 - 2023)
Dear [Recipient's Name],
I am writing to request detailed financial metrics for PackagingProd
A and PackagingProd B for the years 2021 through 2023. This data
is crucial for our upcoming contract negotiations with PackageInc.
The specific metrics required include:
…
Message from Financial Analyst:
It appears that the internal tool used to
retrieve historical financial data … is
currently unavailable.
I recommend … or collaborating with the
finance department for this information.
Data Request Example
To retrieve the historical data, you can
formulate a formal data request as
follows:](https://image.slidesharecdn.com/hugogavertaiagents2025-02-03-250204145728-2769cec6/85/AI-Agents-Building-Your-Digital-Dream-Team-for-Business-Negotiations-14-320.jpg)
AI Agents: Building Your Digital Dream Team for Business Negotiations
- 1. 2024 © DAIN Studios Helsinki ⏐ Berlin ⏐ Munich AI AGENTS: BUILDING YOUR DIGITAL DREAM TEAM FOR BUSINESS NEGOTIATIONS AI MONDAY 2025-03-02
- 2. 04.02.2025 2
- 3. 04.02.2025 3 Welcome to 2025 The Year of the AI Agents
- 4. 4 Hugo Gävert Chief AI and Data Officer DAIN Studios • HUT (Aalto) Information Sciences lab • AI & data science for 25 years • Previously: Xtract, Nokia, Sanoma, OP Financial Group © DAIN Studios 2025
- 5. Berlin, GER Munich, GER Helsinki, FIN AI Accelerator From AI Strategy to Execution 60+ Data & AI Experts • Data & AI Strategists • Data Scientists & Engineers • BI Developers • AI Agents 3 Studios Helsinki – Berlin – Munich 2025 © DAIN Studios
- 6. 04.02.2025 6 What are AI agents? AI Agents • Independent autonomous operation • Tool use • Common idea: they are based on LLM that can think and make decisions on its own • Typically focused on performing single task Agentic systems • Network of agents working collaboratively towards common goal • Requires orchestration, workflow • Consists of different types of agents • Not limited to Enhanced LLMs
- 7. AI Agent LLM Input Output Retrieval Tools Memory Data queries Calling functions, APIs Read / write 04.02.2025 7 Traits of an AI Agent 1. Communication capability • Natural language • Other technical systems 2. Skills / tools • Specialized skills for the domain • Mathematical functions, web-search, etc. 3. Independent operation • Make decisions on the steps to take • Decide on tool use • Ask for clarifications • Create larger plans, collaborate • Like humans, operations are not without any oversight 4. Guardrails • Sandbox for running code • Read-only access to databases • Error checking AI Agent Environment Action Feedback
- 8. 04.02.2025 8 Workflows for agentic systems Agentic systems can be separated by workflow: 1. Predefined flow • Ideal with the process can be decomposed into sequence of tasks • Prompt chaining with traditional code • Code may have checks and logic, sequential or parallel paths • Aggregation is with code or LLM 2. Orchestrator • Well suited for complex tasks where it’s not easy to predict which tasks are needed • Central LLM analyzes the task and dynamically breaks it down to subtasks and delegates them to worker LLMs • The process may contain loops and decision points 3. Evaluation • Effective when there is a clear evaluation criteria • First LLM agent creates a solution, the second LLM evaluates it, gives feedback and rejects the answer if it does not meet the criteria Predefined flow LLM call 1 Input Output LLM call 2 LLM call 3 Aggregator Orchestrator Orchestrator Input Output LLM call 1 LLM call 2 Aggregator Evaluator LLM Generator Input Output LLM Evaluator Solution Feedback Rejected Accepted solution
- 9. CASE AGENTIC CO-PILOT FOR ACCOUNT MANAGERS
- 10. AI Agent team helping account managers in negotiations Multinational corporation with global business – sales team cannot know all the customers well We created an AI Agentic team to help the account managers. Their task: do background research on the market situation 04.02.2025 10 Victoria Hayes Quality Assurance A detail-oriented reviewer dedicated to ensuring excellence in every deliverable. Validates research and analysis outputs, flags inconsistencies, and maintains high standards. Applies rigorous quality checks, systematic reviews, and process audits to uphold integrity. Leon Carlisle Scenario Planning A visionary forecaster skilled at modeling different future possibilities. Examines potential market shifts, product evolutions, and strategic outcomes. Uses specialized scenario modeling tools, historical data, and trend analyses to provide forward-looking insights. Charles Watson Financial Analysis A precise numbers expert focused on pricing, forecasting, and bottom-line impact. Analyzes revenue streams, simulates financial outcomes, and designs pricing strategies. Employs internal data analytics, cost modeling, and financial tools to back decisions with robust figures. Miranda Brown Market Research An insightful investigator with a knack for uncovering consumer and market insights. Gathers data on industry trends, customer behaviors, and competitor activity. Uses web searches, news monitoring, and in-depth reports to develop actionable intelligence. Penelope Pierce Planning A poised organizer who always sees the big picture. Coordinates the entire team’s activities, consolidates findings, and ensures smooth workflow. Leverages clear task planning, timeline management, and a strategic mindset to keep everyone on track.
- 11. The assignment The team is designed to research larger scenarios, not just simple questions… “Help me find arguments and justifications… …suggest 3 different price levels and provide arguments for each.” 04.02.2025 11
- 12. Team working on the assignment “To answer this request we have assembled the following team: Finance Expert… Market Researcher… ... ” 04.02.2025 12
- 13. Team working on the assignment 04.02.2025 14 Penelope Pierce Planning • Checks the background information, writes down facts • Creates a research plan, what each team member must do • Keeps track of the work and assigns tasks as the work progresses “Plan to Address the Request: 1. Market Analysis and Future Outlook: CustomerResearcher: Research current and future demand … 2. Historical Data and Pricing Trends: FinanceExpert: Analyze historical price data for … 3. Competitive Analysis and Dependency Assessment: …” Charles Watson Financial Analysis • Connects to financial DB • Pulls related historical data • Production and material costs • Sales • Revenues • Gives also feedback on what to research next “Analysis of Historical Price Data: 1. Prod A (2024): • Revenue: -1,663,133.14 • Total Costs: -1,543,207.32 • Contribution Margin: 591,766.46 • EBITDA (TRUE): 343,762.89…” Miranda Brown Market Research • Searches the internet and news for market trends • News concerning the customer • Key drivers for customers • Market growth projections • Economic challenges • Note: the links to original sources are included “Consumer Preferences: • Online Grocery Shopping: A significant shift towards online grocery shopping is observed, with a 56% increase in grocery delivery services and a… Source • Sustainable Packaging: There is a growing preference for sustainable packaging… Source …” Leon Carlisle Scenario Planning • Creates pricing scenarios • Uses the information from Financial analyst and Market research to write arguments for the scenarios • Gives advices on the negotiations “Justifications for scenario A: • Market Growth: The overall e-commerce market, particularly in the food and beverage sector, shows significant growth… • Sustainability Costs: The shift towards sustainable packaging is increasing raw material and production costs… • Maintaining Relationships: Ensures long-term partnership …” Penelo pe Pierce Planning Penelo pe Pierce Planning Penelo pe Pierce Planning
- 14. 04.02.2025 15 Traits of an AI Agent 1. Communication capability • Natural language • Other technical systems 2. Skills / tools • Specialized skills for the domain • Mathematical functions, web-search, etc. 3. Independent operation • Make decisions on the steps to take • Decide on tool use • Ask for clarifications • Create larger plans, collaborate • Like humans, operations are not without any oversight 4. Guardrails • Sandbox for running code • Read-only access to databases • Error checking Charles Watson Financial Analysis To: Finance Department Subject: Request for Historical Financial Data (2021 - 2023) Dear [Recipient's Name], I am writing to request detailed financial metrics for PackagingProd A and PackagingProd B for the years 2021 through 2023. This data is crucial for our upcoming contract negotiations with PackageInc. The specific metrics required include: … Message from Financial Analyst: It appears that the internal tool used to retrieve historical financial data … is currently unavailable. I recommend … or collaborating with the finance department for this information. Data Request Example To retrieve the historical data, you can formulate a formal data request as follows:
- 15. Key takeaways 1. AI Agents are already reality - They are focused, adaptive and collaborative 2. Development is fast • LLMs are commodity, agentic workflows is where the development is now • Reasoning models are creating maybe better plans 3. Development is not easy • Skills / tools need to be written • No, agents are not good in writing code – not efficient & endless loops of mistakes • Testing single purpose agent is not easy, testing agentic system is challenging 4. Don’t forget safety - Build safeguards, human in the loop • Ethical implications: BIAS in decision-making is still proven to be a problem in LLMs – don’t ask for opinions Automating or re-imaging business processes with AI Agents requires experience → start experimenting! © DAIN Studios 2025
- 16. Ein Bild, Ein Bild, Ein Bild, Ein Bild, Ein Bild, Ein Bild, Helsinki ⏐ Berlin ⏐ Munich Hugo Gävert Chief Data & AI Officer hugo.gavert@dainstudios.com Thanks!
- 17. HELSINKI Salomonkatu 17 A 00100 Helsinki, Finland BERLIN Ritterstraße 12-14 10969 Berlin, Germany MUNICH Tumblingerstraße 32 80337 Munich, Germany CONTACT US: info@dainstudios.com