Agentizing Business Process

Agentizing Business Process

Feel the AI stones to cross the agentic river

TL;DR

  1. Agentization of business processes has started
  2. Understanding processes and deconstructing tasks is the first step
  3. Agentization has number of hidden assumptions and challenges
  4. Proposed some ideas for successful agentization

In today's rapidly evolving business landscape, organizations are increasingly looking toward AI agents to transform their processes from routine to remarkable. But what does it really mean to "agentize" your process, and how can you navigate this journey successfully?


Understanding Agents

What makes AI agents different from traditional software? There are several characteristics:

  • Human/process-level outcome-driven: Different from traditional software focused on computational outcomes; business model based on impact rather than code volume/complexity
  • Non-determinism: Many possible outcomes based on context/need (software with reasoning); same question may yield alternate (and better) answers
  • Autonomy: Automated decision making; steering instead of controlling
  • Automated Code Generation: Most code of agents is expected to be generated by LLMs.
  • Near-infinite scale: Endless parallel operations; endlessly evolving business - tomorrow's business is not today's business

As a result of these, the structure of software industry is going through massive change. Agents are not just about more software. It is about different way of doing business - one that is more knowledge, skill, and impact centered as opposed to code centered.


Understanding Business Processes

Traditional organizational processes were designed with specific ideas:

  • Human Scaling: Designed around roles and organizational boundaries, for monitoring and managing trust and quality, and for managing evolution over time
  • Task Hierarchy: Business processes have well-defined task hierarchy based on skill and complexity required. This is because of limited talent pool. Companies like Google had 7+ levels of staff. Commonly we have L1 = Well defined narrow tasks, L2 = Domain-specific complicated tasks, and L3+ = Expert-level difficult tasks
  • Stability : Organizations are long-lived entities, and the processes survive for a long period of time. The mental models around organizational structures assume long timeframes.


Agentization requires fundamentally different assumptions and will ultimately rewire the organization. The lifetime of products and services will be short, the availability of high skill will be unconstrained (other than by money), and evolution will be fast.


Challenges in Agentization

A process or task can be agentized only under certain conditions. There are a number of challenges but mainly along two dimensions:

Organizational:

  • Human verification: Completeness, accuracy, legality etc.; default mode, not risk driven
  • Coordination: Approvals, inputs from various stakeholders
  • Ambiguity: Boundaries, assignment, skill levels that aren't clearly defined
  • Risk: Financial, health, legal implications that require careful management

Technological:

  • Model limitations: Media handling, knowledge gaps in specialized domains
  • Decision complexity: Infinite choice spaces; fast evolution of technology and requirements
  • Infrastructure: Data quality & instrumentation needs for effective agent operation
  • Skill: Non-deterministic software development requiring new competencies

Understanding tasks and matching it to capability of the current technology is critical.


Approaching Process Agentization

The best candidates for agentization exist at the intersection of high value and high volume, with a focus on verifiable tasks. Tasks that can be objectively evaluated tend to make better candidates for automation through agents.


  • Start with realistic expectations: End-to-end agentization may/may not be possible; consider a Ship of Theseus approach to gradual transformation
  • Choose strategic starting points: Value-driven selection; Theory of constraints or other methodologies to identify bottlenecks
  • Redefine task boundaries to make them AI-friendly and more suitable for agent processing
  • Build robust solutions: From simple implementations to full AI stack including UX; design for robustness with respect to technological evolution


Recommendations for Success

  • Take a process view rather than focusing solely on individual tasks
  • Co-develop change management plan: Take people along with a graceful path including upskilling and new initiatives/careers; find champions within the organization
  • Use Theory of Constraints or other means to identify candidates for maximum impact
  • Pick high value verifiable tasks first: Choose simplest techniques that will be robust with future changes
  • Do one and scale it up, then move to the next rather than trying to transform everything at once
  • Define next gen Agent HR Ops division: Develop methodology, skillsets, and process adaptation frameworks


The journey isn't just about implementing technology—it's about reimagining how work gets done. The organizations that will thrive are those that approach this transformation thoughtfully, focusing on both the technological and human dimensions of change.

What processes in your organization are ripe for agentization? I'd love to hear your thoughts in the comments below.

#AIAgents #ProcessTransformation #DigitalTransformation #FutureOfWork #AgentizingProcess

Jessica Jones

Doing Something Great | Growth Leader | Speaker | Ex-Google

1 天前

Venkata Pingali Agentization feels like a step beyond just streamlining workflows...it’s rethinking *why* we do things the way we do. I’ve found that asking ‘What assumptions are baked into this process?’ can uncover surprising opportunities to simplify or innovate.

回复

要查看或添加评论,请登录

Venkata Pingali的更多文章

  • Agent-Based Systems Have Arrived: AI Engineer Summit Online 2025

    Agent-Based Systems Have Arrived: AI Engineer Summit Online 2025

    TL;DR: The AI Engineer Online Summit 2025 shows that AI agents are rapidly maturing. The talks had a strong sense of…

    10 条评论
  • Where will LLMs be in the Next 12 Months?

    Where will LLMs be in the Next 12 Months?

    Benchmarks. Normally we like to think of technology development as an independent process dictated by markets.

  • Agents Will Take Over IT Service Management

    Agents Will Take Over IT Service Management

    TL;DR ITSM economics is about to breakdown ITSM has a long tail of use cases because of complexity Agents will be…

    1 条评论
  • [Feb 5] Implementation Experiences with Domain LLMs

    [Feb 5] Implementation Experiences with Domain LLMs

    A lot of theoretical work is happening but delivering it to end customers is still a bit of challenge. This week we…

  • Post-Deepseek World

    Post-Deepseek World

    Deepseek has reset priors of the tech community at large, and opened a much larger application game. Here is a mix of…

    4 条评论
  • Jan 24, 2025 - Knowledge Agents & Economics

    Jan 24, 2025 - Knowledge Agents & Economics

    Welcome! In this edition we have two articles written by me and Rajesh on structure of knowledge agents, and economics…

  • Alignment is Critical: What I’ve Learned About Leading a Cross-Border Startup

    Alignment is Critical: What I’ve Learned About Leading a Cross-Border Startup

    Leading a cross-border organization has taught me that success depends on understanding and adapting to unique…

    6 条评论
  • A Year to Remember

    A Year to Remember

    It was the best of times, it was the worst of times, it was the age of wisdom, it was the age of foolishness, it was…

    5 条评论
  • A CEO’s Guide: The Brave New World of Data Privacy and Accountability

    A CEO’s Guide: The Brave New World of Data Privacy and Accountability

    The compliance landscape involving data is significantly changing in 2020, and it is necessary to understand these…

    1 条评论
  • CEO Guide to Production ML

    CEO Guide to Production ML

    Productionization or operationalization of Machine Learning is the process of making machine learning models run every…

    3 条评论