# ai agents transforming enterprises ![[cover.png]] denys holovatyi, founder & ceo @ osnova gmbh --- # contents - [[#what is artificial intelligence?]] - [[#can ai think?]] - [[#ai in organizations]] - [[#pitfalls of ai]] - [[#contact]] - [[#hiring]] --- # what is artificial intelligence? ## classical definition a research area of computer science that deals with the simulation of human intelligence in computers ai researches self-learning systems that recognize patterns in the training data and are not explicitly programmed ![[black-and-white-photo-of-seven-smiling-men-sitting-on-a-lawn-in-front-of-a-tree-and-a-white-school-building-with-many-windows.webp]] [dartmouth workshop](https://en.wikipedia.org/wiki/dartmouth_workshop), 1956, organized by [john mccarthy](https://amturing.acm.org/award_winners/mccarthy_1118322.cfm), [marvin minsky](https://en.wikipedia.org/wiki/marvin_minsky), and [nathaniel rochester](https://en.wikipedia.org/wiki/nathaniel_rochester_(computer_scientist)), and [claude shannon](https://spectrum.ieee.org/claude-shannon-tinkerer-prankster-and-father-of-information-theory) (founder of information theory) --- ## behavior-based definition the ability of a machine to imitate human behavior --- ## let us ask the ai itself ```instruction What is artificial intelligence? ``` ![[2025-04-06 14-35-34.mkv]] --- # can ai think? the question of whether artificial intelligence can think has been disputed for decades. in the last two years more and more researchers believe that ai shows various advanced cognitive capabilities. for example, it is able to think about human instructions instead of executing them directly. ![[2025-04-06 14-49-18.mkv]] --- ## assistant vs reasoner vs agent ### assistant an ai model that is focused on user interaction and supports humans with requests. it is trained to provide helpful and honest answers, provide information, and communicate in a dialogue-oriented format. ### reasoning model an ai model specializing in logical thinking and problem solving. it can perform structured analyses, develop step-by-step solutions, and draw complex conclusions, often using chain-of-thought methods. ### agent an autonomous ai system that can act independently, make decisions, and perform tasks. agents interact with their environment, can access external tools, pursue specific goals, and independently plan complex task sequences. --- ## ai agents of the future are here ![[dbc0287f-a878-42cb-8db2-f84104470413_1132x805.webp]] [link](https://www.latent.space/p/agent) --- ### manus agent creates 50 twitter accounts simultaneously ![[_SIVqaoinmaI1Ke7.mp4]] --- ### claude agent generates 3d objects in blender with mcp integration ![[MCP_3D建模的一个用例可以让Claude直接与Blender对话的MCP：blender_mcp.mp4]] --- ### perplexity agent researches artificial intelligence ![[2025-04-06 15-03-01.mkv]] --- # ai in organizations ## development of company-specific ai with internal knowledge companies are increasingly asking how they can integrate advanced intelligence into their workflows. when considering intelligent search solutions, many organizations imagine loading all company data into artificial intelligence - connecting sharepoint, confluence, documents and other sources to create a comprehensive knowledge base --- ## increasing productivity through process automation with ai in addition to intelligent search with chatbots, companies are interested in process automation with ai modern language models like chatgpt can automate various process activities, as these often represent nothing more than moving documents from a to b and processing them. most documents - such as orders, patents, or resumes - are text-based and can thus be evaluated and generated by language models with high quality --- ## automated order processing automated order processing works as follows: - orders or customer orders are read from the outlook inbox - data from the orders are extracted in raw format - raw data is transformed into a standardized format - the new data object is written back to the erp system the order managers can thus save time in order processing. the advantage of ai is that it can understand different templates of orders, which is impossible for workflow automation or a batch job. osnova has implemented such ai projects for several customers, especially for medium-sized manufacturing companies ![[Pasted image 20241112151101.png]] --- ## patent analysis with prompt sequences patent analysis is a very complex process. it requires both legal and engineering knowledge. a patent analysis can take 4 hours, but can be largely automated with ai - analysis time is reduced to 15 minutes, and even without human intervention. for this purpose, a sequence of commands to the ai, called "prompt sequence", must be developed. in each step, a specific aspect of the patent analysis is carried out, information is compared with technical standards and iteratively validated until the quality of the results is correct. - highly complex **prompt sequence** of 13 prompts and 19 versions until the corresponding quality level was achieved - analysis of 50,000 patents - reduction of analysis time from 3-4 hours to 15 minutes this project was carried out for an innovative patent law firm --- ## evaluation of resumes with ai evaluation of resumes is an important task, but most recruiters only have 6 seconds for it an ai can evaluate a resume in detail according to the company's criteria and the requirements of the job advertisement and offer the recruiter a detailed basis for decision-making, and if needed, it can also generate interview questions this is particularly helpful for highly specialized positions where the recruiter has no technical expertise but needs to quickly identify high-caliber applicants ![[Pasted image 20250415172514.png]] --- ## ai use cases you can see more below. use cases are in german, but you can easily use llms to translate - [[use cases - bank]] - [[use cases - insurance]] - [[use cases - engineering services]] - [[use cases - metal products manufacturer]] --- ## llm prompts start here: - [[prompt - cv questions]] - [[prompt - company research]] - [[bot - @prompter-3000]] - [[bot - @botmaker]] you can use @prompter and @botmaker to create your own prompts in seconds 1. @prompter 1. [on Poe](https://poe.com/prompter-3000x335) 2. [on ChatGPT](https://chatgpt.com/g/g-67f2765cbae4819194179538e5e6f860-prompter-3000x335) 2. @botmaker 1. [on Poe](https://poe.com/botmaker-0000) 2. [on ChatGPT](https://chatgpt.com/g/g-67f287170f008191b3a743ecf95068e8-botmaker) ![[Pasted image 20250416122222.png]] --- # pitfalls of ai ## technological ### ai hallucinations ai hallucinations are well-known and common occurrences where language models provide factually incorrect information with confidence. while these are considered mistakes, they're actually inherent to the stochastic nature of language models. every generated text is technically a hallucination, though most are useful quality can be improved by: - supplying contextual information in queries - using retrieval augmented generation (RAG) to pull relevant text from company databases --- ### context recall problems ai systems struggle with memory limitations. even when provided with extensive content (like the entire harry potter series), they forget significant portions, leading to imprecise or incorrect answers context recall: - begins declining after just 2,000 tokens - "completely plummets" beyond 32,000 tokens --- ### reliability challenges the reliability of ai agents remains an unsolved problem even with a 90% success rate at individual steps, when connecting multiple steps (e.g., 20), the probability of complete success drops dramatically to approximately 12% this explains why tools like deep research cannot operate autonomously and require precise engineering --- ## economic, social, and political ### job displacement while technology leaders discuss job creation and education to counteract ai-driven job displacement: - in practice, insufficient resources are allocated to address this issue - current estimates suggest hundreds of millions of jobs could be affected - the pace of change far exceeds that of the industrial revolution, leaving workers minimal time to adapt - this rapid transformation intensifies current and future inequality --- ### market concentration as of april 2025, the ai market is highly concentrated and oligopolistic: - only a handful of companies possess frontier-level models - other organizations must purchase ai capabilities from these dominant providers - as ai becomes business-critical rather than optional, power and wealth will increasingly flow toward american technology giants --- ### intellectual property concerns ![[Pasted image 20250415200833.png]] [link](https://x.com/jack/status/1910829254214115681) ip protection is becoming a contentious issue: - some tech leaders like jack dorsey and elo musk advocate abolishing ip laws altogether - companies outside american big tech may need to advocate for stronger ip protections - current enforcement appears uneven, with major tech companies able to use copyrighted materials with limited consequences --- # contact denys holovatyi founder & ceo osnova gmbh [linkedin](https://www.linkedin.com/in/denysholovatyi/) [email protected] --- # hiring ## intro i am making a switch from ai consulting to building a product cogit is an operating system for ai agents personal, owned, self-learning looking for people to help me with funding & re-igniting my network ## roles looking for interns - funding search - find funding programs - help prepare applications - business development - outreach on linkedin, lead research, email direct - preparation of events with ai founders & developers --- ## send your cv and 5 bullet points about how you would tackle the problems of funding research and business development --- --end--