@victordibia.bsky.social
Principal RSDE, @Microsoft focusing on #GenAI Agents, #pyautogen | ex @cloudera @ibmresearch, Created #lida . Tweets my own. http://victordibia.com
It’s finally here! Just got an author proof copy of the designing multi-agent systems book. It’s taken over a year to get this done, and so it’s quite exciting to see it wrapping up!
Digital: buy.multiagentbook.com
Print : - ETA Nov 15 on Amazon
Digital: buy.multiagentbook.com
Print : - ETA Nov 15 on Amazon
October 29, 2025 at 4:04 PM
It’s finally here! Just got an author proof copy of the designing multi-agent systems book. It’s taken over a year to get this done, and so it’s quite exciting to see it wrapping up!
Digital: buy.multiagentbook.com
Print : - ETA Nov 15 on Amazon
Digital: buy.multiagentbook.com
Print : - ETA Nov 15 on Amazon
It's taken a while .. but something's cooking ...
if you have enjoyed any of my previous writing, I think you'd find this useful too.
if you have enjoyed any of my previous writing, I think you'd find this useful too.
September 22, 2025 at 3:47 PM
It's taken a while .. but something's cooking ...
if you have enjoyed any of my previous writing, I think you'd find this useful too.
if you have enjoyed any of my previous writing, I think you'd find this useful too.
Can You Build Agent2Agent Communication on MCP?
Yes.
I wrote an article about how you can compose some of the recent features within MCP( resumable streams, elicitation, sampling, and notifications ) to implement long running agents (as tools):
devblogs.microsoft.com/blog/can-you...
Yes.
I wrote an article about how you can compose some of the recent features within MCP( resumable streams, elicitation, sampling, and notifications ) to implement long running agents (as tools):
devblogs.microsoft.com/blog/can-you...
July 24, 2025 at 10:25 PM
Can You Build Agent2Agent Communication on MCP?
Yes.
I wrote an article about how you can compose some of the recent features within MCP( resumable streams, elicitation, sampling, and notifications ) to implement long running agents (as tools):
devblogs.microsoft.com/blog/can-you...
Yes.
I wrote an article about how you can compose some of the recent features within MCP( resumable streams, elicitation, sampling, and notifications ) to implement long running agents (as tools):
devblogs.microsoft.com/blog/can-you...
4 UX Design Principles for Semi-Autonomous Multi-Agent Systems
I gave a keynote talk at the June AI.Engineer conf
newsletter.victordibia.com/p/4-ux-desig...
Bonus: I cover lessons from building a multi-agent system from scratch
I gave a keynote talk at the June AI.Engineer conf
newsletter.victordibia.com/p/4-ux-desig...
Bonus: I cover lessons from building a multi-agent system from scratch
July 14, 2025 at 7:02 PM
4 UX Design Principles for Semi-Autonomous Multi-Agent Systems
I gave a keynote talk at the June AI.Engineer conf
newsletter.victordibia.com/p/4-ux-desig...
Bonus: I cover lessons from building a multi-agent system from scratch
I gave a keynote talk at the June AI.Engineer conf
newsletter.victordibia.com/p/4-ux-desig...
Bonus: I cover lessons from building a multi-agent system from scratch
PSA: Start with a non-agent baseline, invest in tools and evals, before implementing complex agentic behaviours newsletter.victordibia.com/p/4-ux-desig...
July 14, 2025 at 6:59 PM
PSA: Start with a non-agent baseline, invest in tools and evals, before implementing complex agentic behaviours newsletter.victordibia.com/p/4-ux-desig...
Some of the hardest posts to write are not the ones that require writing code or running extensive experiments. Its usually the ones that require sifting through a host of noisy ideas, brutally clear thinking and distilling clear insights or positions.
January 18, 2025 at 10:52 PM
Some of the hardest posts to write are not the ones that require writing code or running extensive experiments. Its usually the ones that require sifting through a host of noisy ideas, brutally clear thinking and distilling clear insights or positions.
An interesting conundrum with working with AI tools (e.g., a coding assistant in this case) is that you can genuinely use it to create a complex system which you do not understand.
January 15, 2025 at 9:23 PM
An interesting conundrum with working with AI tools (e.g., a coding assistant in this case) is that you can genuinely use it to create a complex system which you do not understand.
A the AI field evolves, the value of the ML/AI PhD is changing newsletter.victordibia.com/p/you-have-a...
January 9, 2025 at 9:29 PM
A the AI field evolves, the value of the ML/AI PhD is changing newsletter.victordibia.com/p/you-have-a...
The lesson?
Right now is the greatest time to work on a side project.
Right now is the greatest time to work on a side project.
January 5, 2025 at 5:12 AM
The lesson?
Right now is the greatest time to work on a side project.
Right now is the greatest time to work on a side project.
5. End-to-End Agent Benchmarks
Before: Ad-hoc evaluation focused on individual model capabilities (language, reasoning, tool use)
Trend: Comprehensive task completion benchmarks measuring end-to-end agent performance.
Before: Ad-hoc evaluation focused on individual model capabilities (language, reasoning, tool use)
Trend: Comprehensive task completion benchmarks measuring end-to-end agent performance.
January 3, 2025 at 4:10 PM
5. End-to-End Agent Benchmarks
Before: Ad-hoc evaluation focused on individual model capabilities (language, reasoning, tool use)
Trend: Comprehensive task completion benchmarks measuring end-to-end agent performance.
Before: Ad-hoc evaluation focused on individual model capabilities (language, reasoning, tool use)
Trend: Comprehensive task completion benchmarks measuring end-to-end agent performance.
4. A Shift to Complex Tasks and the Rise of Frameworks
Before: Agents built as chains and tool-calling sequences patterns applied to simple tasks.
Trend: Sophisticated patterns for handling complex, multi-step tasks requiring planning, reflection, and coordination.
Before: Agents built as chains and tool-calling sequences patterns applied to simple tasks.
Trend: Sophisticated patterns for handling complex, multi-step tasks requiring planning, reflection, and coordination.
January 3, 2025 at 4:10 PM
4. A Shift to Complex Tasks and the Rise of Frameworks
Before: Agents built as chains and tool-calling sequences patterns applied to simple tasks.
Trend: Sophisticated patterns for handling complex, multi-step tasks requiring planning, reflection, and coordination.
Before: Agents built as chains and tool-calling sequences patterns applied to simple tasks.
Trend: Sophisticated patterns for handling complex, multi-step tasks requiring planning, reflection, and coordination.
3. Interface Agents Take Center Stage
Before: The action space of agents was typically from programmatic tools and code execution
Trend: Direct manipulation of user interfaces (web, desktop) as the primary method for agent action.
Before: The action space of agents was typically from programmatic tools and code execution
Trend: Direct manipulation of user interfaces (web, desktop) as the primary method for agent action.
January 3, 2025 at 4:10 PM
3. Interface Agents Take Center Stage
Before: The action space of agents was typically from programmatic tools and code execution
Trend: Direct manipulation of user interfaces (web, desktop) as the primary method for agent action.
Before: The action space of agents was typically from programmatic tools and code execution
Trend: Direct manipulation of user interfaces (web, desktop) as the primary method for agent action.
2. “Agent-Native” Foundation Models.
The important trend here is that these agentic capabilities are now being “lifted” into the generative model itself.
The important trend here is that these agentic capabilities are now being “lifted” into the generative model itself.
January 3, 2025 at 4:10 PM
2. “Agent-Native” Foundation Models.
The important trend here is that these agentic capabilities are now being “lifted” into the generative model itself.
The important trend here is that these agentic capabilities are now being “lifted” into the generative model itself.
1. Many enterprises and startups have adopted the term “agents” in products, describing them broadly at systems that act on a user’s behalf with the goal of saving the user time and avoiding tedious/busy work
January 3, 2025 at 4:10 PM
1. Many enterprises and startups have adopted the term “agents” in products, describing them broadly at systems that act on a user’s behalf with the goal of saving the user time and avoiding tedious/busy work
AI Agents 2024 Rewind - A Year of Building and Learning
What were the top 5 trends in AI Multi-Agent Systems from 2024? What can we expect in 2025?
newsletter.victordibia.com/p/ai-agents-...
What were the top 5 trends in AI Multi-Agent Systems from 2024? What can we expect in 2025?
newsletter.victordibia.com/p/ai-agents-...
January 3, 2025 at 4:10 PM
AI Agents 2024 Rewind - A Year of Building and Learning
What were the top 5 trends in AI Multi-Agent Systems from 2024? What can we expect in 2025?
newsletter.victordibia.com/p/ai-agents-...
What were the top 5 trends in AI Multi-Agent Systems from 2024? What can we expect in 2025?
newsletter.victordibia.com/p/ai-agents-...
One of the many things I have been reminded of this year while working on AutoGen and AutoGen Studio is creating the right MVP.
December 19, 2024 at 5:26 PM
One of the many things I have been reminded of this year while working on AutoGen and AutoGen Studio is creating the right MVP.
References
[1] newsletter.victordibia.com/p/machine-le...
[2] Peacasso github.com/victordibia/...
[3] Whisk - labs.google/fx/tools/whisk
newsletter.victordibia.com/p/machine-le...
[1] newsletter.victordibia.com/p/machine-le...
[2] Peacasso github.com/victordibia/...
[3] Whisk - labs.google/fx/tools/whisk
newsletter.victordibia.com/p/machine-le...
December 18, 2024 at 6:17 PM
References
[1] newsletter.victordibia.com/p/machine-le...
[2] Peacasso github.com/victordibia/...
[3] Whisk - labs.google/fx/tools/whisk
newsletter.victordibia.com/p/machine-le...
[1] newsletter.victordibia.com/p/machine-le...
[2] Peacasso github.com/victordibia/...
[3] Whisk - labs.google/fx/tools/whisk
newsletter.victordibia.com/p/machine-le...
Models that generate high quality visualizations have been available for a while .. but I still get amazed from time to by by tools that tie together user experiences that enable human creativity …
Pretty cool tool from Google Labs - Whisk. (h/t to Alex Chen and team!)
labs.google/fx/tools/whisk
Pretty cool tool from Google Labs - Whisk. (h/t to Alex Chen and team!)
labs.google/fx/tools/whisk
December 18, 2024 at 6:17 PM
Models that generate high quality visualizations have been available for a while .. but I still get amazed from time to by by tools that tie together user experiences that enable human creativity …
Pretty cool tool from Google Labs - Whisk. (h/t to Alex Chen and team!)
labs.google/fx/tools/whisk
Pretty cool tool from Google Labs - Whisk. (h/t to Alex Chen and team!)
labs.google/fx/tools/whisk
December 14, 2024 at 10:45 PM
Just as expertise in traditional programming languages, integrating task-specific machine learning models were the critical for great engineers, I increasingly believe being able to build multi-agent pipelines will offer similar differentiation.
December 13, 2024 at 6:36 AM
Just as expertise in traditional programming languages, integrating task-specific machine learning models were the critical for great engineers, I increasingly believe being able to build multi-agent pipelines will offer similar differentiation.
On The Critical Role of Verification in Human-AI Agent Interaction
In this video from [1], they demonstrate the new 01 preview model solving a complex thermodynamics problem involving surface area calculations for a cooling radiator panel, for a data center in space.
In this video from [1], they demonstrate the new 01 preview model solving a complex thermodynamics problem involving surface area calculations for a cooling radiator panel, for a data center in space.
December 12, 2024 at 5:23 PM
On The Critical Role of Verification in Human-AI Agent Interaction
In this video from [1], they demonstrate the new 01 preview model solving a complex thermodynamics problem involving surface area calculations for a cooling radiator panel, for a data center in space.
In this video from [1], they demonstrate the new 01 preview model solving a complex thermodynamics problem involving surface area calculations for a cooling radiator panel, for a data center in space.
It **should** be easy to define multi-agent workflows declaratively, build them from composable components (agents, termination conditions, models, tools etc) and test/debug them.
December 11, 2024 at 9:18 PM
It **should** be easy to define multi-agent workflows declaratively, build them from composable components (agents, termination conditions, models, tools etc) and test/debug them.