As we unleash new & unique powers for our end-users and business teams supporting them this quarter, I wanted to share the vision of the platform.
See also our point of view on AI agent session replay here.
Intelligent. Simple. Integrated. Open. Flexible. Hybrid. Secure.
IBM watsonx Orchestrate (wxO) is to be the orchestrator of all agents, meeting the users where they are, allowing to
-
tap into any existing IT investment (biz app, models, agents)
-
fill the gaps in no/low-code allowing non-technical teams to configure/build from scratch, with the help of AI or from pre-built agents
-
accelerate time-to-value for end-user, offering a modern conversational UI to simplify their life, hiding the complexity and variability of the IT stack supporting their work.
wxO is…
Intelligent.
wxO provides OOTB an LLM-driven orchestrator of agents to allow any type of agent (& assistant) to collaborate seamlessly to serve the end-user.
Agents can be guided to collaborate with each other through instructions given in natural language by the no-code builder. Some say, it is like talking to a 5 year old; lay it out for them in plain English (and soon other languages) terms and it will follow the instructions!
Think of simple generic things like “whenever the response includes a list of more than 3 items, display as a table“ or more business knowledge such as “if given a User ID, look up the user name and email in tool X“. =
This essentially teaches upfront the LLM to (better) serve the end-user. We can imagine in the future the end-user being able to add their own instructions so the LLM adapts to their needs & preferences (eg “only display as a table when more than 5 items in the list, otherwise a text list is ok”).
Agents can also be given more specific / structured instructions in the form of
-
business rules loaded in the form of spreadsheets to codify specific (& regulated) business policies
-
flows drawn visually in a low-code graphical editor, allowing to define procedures as a specific sequence of steps that can be conditioned to take various paths based on the context.
It all together makes "building” pretty intuitive and accessible!
Simple.
wxO core value is reducing/managing the complexity of a growing agent landscape & proliferation of solutions already invested in/used (like with apps before) so the end-user does not have to carry that burden of knowing where to go to do what.
It also provides a no-code agent builder UI to allow non-technical folks to bring all of that together in a seamless manner, with the option to import existing agents built by other teams using other techs/vendors and/or configure their own:
-
you don't need to look for an agent builder if you need one!
-
it is no-code, so very much accessible to non-tech (unlike most agent builders out there)
-
less/little to build, just
-
describe your agent purpose or load pre-built agents part of our catalog (HR, Sales, Procurement & Customer Care coming soon)
-
configure your agents with
-
tools (see below)
-
knowledge (data/docs/website for RAG)
-
collaborators (other agents to interact with)
-
guidance (instruction telling it what to do and how in natural language)
-
then let the LLM do the (hard) work of orchestrating based on user’s asks!
The agent builder provides all ones need to address the user’s need so the agent can act on their behalf across any system:
Integrated. Open. Flexible.
It is not all or nothing with IBM: we have it all but does not close you into its own tech: use what you have, it's an "à la carte" style of platform!
One of the key motto and value is to allow to tap into existing investments (new & old tech) in agents & tools, like it already did/does with assistants & skills. At IBM, we know very well that organization often have a plethora of techs & vendors various groups are using and a history of IT investments that needs to be taken into account as technology evolves constantly but moving off the legacy tech can take time.
We do so using the leading industry standards, starting with chat completion API & OpenAPI , and looking at other interface standards as those possibilities continuously evolves & expand as well.
Also a key aspect is wxO is ultimately designed to serve the end-user so it ought to meet they where they already are
-
provide a responsive / mobile-friendly & embeddedable / brandable web UI adapted to their device & org look & feel
-
allow to be reach out to through mainstream com tools such as Slack, Teams, WhatApps or other agent tech (eg Co-pilot, Agent Force…)
The bottom line is that if the user spends a majority of their time in a given app, the agent should meet them there, regardless the vendor (IBM or not). As the list of apps is never ending and always changing, one future possibility would be a plugin for the web browser which is definitely a key common denominator in everyone’s day at work.
LLM can now handle most languages, therefore so will an AI agent: once given proper instructions, it will answer in the language being addressed in!
wxO is very focused on no-code, unlike some of they key early agentic techs, the aim being to finally democratize the building of business value and allow lines of business to handle some of their their own. That being say wxO does not close our customers to only one way to do things as the orchestrator of agent allows to bring together:
-
LLM-power agents & legacy assistants (wxO, wxA & beyond)
-
wxO agents & non-wxO agents (eg wx.ai)
-
IBM agents & non-IBM agents (eg MS Copilot or SF AgentForce)
-
no-code, low-code & pro-code agents
IBM definitely stays true to its hydrid story, knowing that our customers have to juggle with various needs & constraints in terms of IT architecture and deployments:
-
Onprem & SaaS
-
New & legacy tech
-
IBM & non-IBM tech
wxO allow to bring all of the IT stack together with also CI/CD enabled.
For those wanting to go deeper / more custom, we will enable to run python code within the platform.
Secure
wxO is grounded into the IBM culture on security and trust. It is embedded in to the AI technology of our own wx.ai platform which is underlying any LLM feature in wxO; be it the builder configuring tools and agents, or the end-user at runtime. Governance is infused at the platform, agent and tool level. The latter leverages credentials from the corresponding apps to define access rights to any data made accessible there, so a wxO user will always only be allowed to do what they could do in the corresponding 3rd-party app that wxO is acting on on their behalf. It is making the work faster but not less secure!
Even more so as access to given agents and the platform itself can be controlled as well, at the team/role/group and individual level. As we eventually allow customers to bring their own model through wx.ai, they will be able to tap into the full power of wx.gov & wx.data in order to control the full stack serving the customer, grounded in their knowledge and using their know-how.
In closing/summary
Ultimately wxO brings together the best of IBM in
-
AI (infused at every level; piece of no/low-code wx.ai in the box!)
-
integration (large/growing catalog of SSO-powered connectors)
-
automation (best in class process, decision, IDP & RPA engines)
-
security / governance on models, data and all assets
-
openeness: BYO-model/agent/tool/UI...
but also the best of the whole IT ecosystem by allowing to also tap into all existing (non-IBM) investments.
The wxO promise is
-
world-class end-user experience
-
low-tech builder empowerment
-
short time-to-value & low cost-of-ownership
-
trust on biz outcome & data integrity
-
value existing investments & fill the gaps
Looking forward to hear your feedback & successes, get started at https://www.ibm.com/products/watsonx-orchestrate & let us know what else wxO can do at https://watson-orchestrate.ideas.ibm.com/