This discussion around the 'Not Missing the AI Train' proposal has been insightful, and I appreciate the proactive decision to pause in light of Offchain Labs' existing efforts. This reflects a key strength of decentralized governance, adaptive decision-making based on community feedback and alignment with ongoing initiatives.

I am a little worried however that such a large chain/project has solutions being modelled out for such a large industry by only one team, I think there is still room for a project like this, perhaps on a smaller scale with key deliverables such as 'collaborate to launch an AI Agent launch pad' create a weekly or monthly demo show on youtube, monthly hackathon collaboration with prizes.

Smaller team, smaller budget. AI is not a flash in the pan, every single one of the top tech companies has almost dropped everything they're doing to also 'jump on the bandwagon', we need to not only jump, but LEAP!

I'm really looking forward to seeing how this progresses.

This discussion around the 'Not Missing the AI Train' proposal has been insightful, and I appreciate the proactive decision to pause in light of Offchain Labs' existing efforts. This reflects a key strength of decentralized governance, adaptive decision-making based on community feedback and alignment with ongoing initiatives.

I am a little worried however that such a large chain/project has solutions being modelled out for such a large industry by only one team, I think there is still room for a project like this, perhaps on a smaller scale with key deliverables such as 'collaborate to launch an AI Agent launch pad' create a weekly or monthly demo show on youtube, monthly hackathon collaboration with prizes.

Smaller team, smaller budget. AI is not a flash in the pan, every single one of the top tech companies has almost dropped everything they're doing to also 'jump on the bandwagon', we need to not only jump, but LEAP!

I'm really looking forward to seeing how this progresses.

Great to see this proposal @danielo! I think that Stylus certainly offers some really unique possibilities for on-chain AI agents. I am no longer at the Arbitrum Foundation, but recently developed a demo and workshop specifically on this topic:

https://github.com/hammertoe/ArbitrumOnchainAgent

https://x.com/HammerToe/status/1887966250191532218

Which got a lot of great response:

https://x.com/diopfode/status/1886774429306511361

And also a very good route to bring devs over to Arbitrum from other ecosystems e.g.:

https://x.com/ajwarner90/status/1886775697911304215

I actually was coming here to the forum to propose something very similar to what you have proposed -- some kind of AI-agent focussed team that can move more rapidly than the Foundation can. It would be great to join efforts. I understand the concerns put forth here by people on the details of the proposal, but as you have pointed out, part of the reason for that is it is such a rapidly developing topic.

@karpatkey I understand your concerns about risk of overlapping with the Foundation. Whilst they have dedicated teams for ecosystem growth and BD, they don't really have much/any DevRel capacity and certainly no-one focussing on this specific topic of AI.

Great to see this proposal @danielo! I think that Stylus certainly offers some really unique possibilities for on-chain AI agents. I am no longer at the Arbitrum Foundation, but recently developed a demo and workshop specifically on this topic:

https://github.com/hammertoe/ArbitrumOnchainAgent

https://x.com/HammerToe/status/1887966250191532218

Which got a lot of great response:

https://x.com/diopfode/status/1886774429306511361

And also a very good route to bring devs over to Arbitrum from other ecosystems e.g.:

https://x.com/ajwarner90/status/1886775697911304215

I actually was coming here to the forum to propose something very similar to what you have proposed -- some kind of AI-agent focussed team that can move more rapidly than the Foundation can. It would be great to join efforts. I understand the concerns put forth here by people on the details of the proposal, but as you have pointed out, part of the reason for that is it is such a rapidly developing topic.

@karpatkey I understand your concerns about risk of overlapping with the Foundation. Whilst they have dedicated teams for ecosystem growth and BD, they don't really have much/any DevRel capacity and certainly no-one focussing on this specific topic of AI.

The following reflects the views of GMX’s Governance Committee, and is based on the combined research, evaluation, consensus, and ideation of various committee members.

---

We support exploring AI within Arbitrum but believe this proposal needs a clearer framework before moving forward.

• Objectives & KPIs: The proposal lacks defined goals and success metrics. A structured plan with measurable outcomes is essential.
• Methodology: Flexibility is fine, but success must be clearly defined—what are the objectives, and how will progress be measured?
• Integration with Existing Initiatives: How will this align with Questbook and SOS? Who will manage collaboration, and do PMs/DAs have the capacity for this? We recommend waiting for SOS to help set funding priorities.
• Strategic Approach vs. Hype: Reacting late to trends isn’t ideal. If AI is a real priority, should ARDC conduct exploratory research first, or should we wait for SOS to guide the DAO’s investment strategy?

We encourage further refinement to ensure this initiative is well-structured, strategically aligned, and properly resourced.

We find the underlying premise of this proposal to be strategically compelling - positioning Arbitrum as a key infrastructure layer for AI agent coordination represents a meaningful opportunity that warrants focused attention and resources.

However, while the proposal effectively identifies the strategic importance of this initiative, in our view, its execution framework requires further refinement and solidification.

The core challenge lies in the proposal's lack of concrete implementation specifics:

• The team's operational approach and specific deliverables remain largely undefined
• The interaction model with existing Arbitrum entities (Foundation, OCL) needs clearer delineation to avoid duplicative efforts
• The absence of defined KPIs and success metrics makes it difficult to evaluate effectiveness

The agile approach to development makes sense given the rapidly evolving nature of AI technology. However, this shouldn't preclude establishing clear initial priorities and concrete success metrics.

We believe this proposal addresses a crucial strategic opportunity but would benefit significantly from a more detailed implementation roadmap, specific initial focus areas and deliverables, clear success metrics and KPIs and explicit coordination mechanisms with existing Arbitrum entities.

We would be happy to support a refined version of this proposal that addresses these elements, as ensuring Arbitrum's positioning in the AI infrastructure space represents a meaningful opportunity.

The following reflects the views of GMX’s Governance Committee, and is based on the combined research, evaluation, consensus, and ideation of various committee members.

---

We support exploring AI within Arbitrum but believe this proposal needs a clearer framework before moving forward.

• Objectives & KPIs: The proposal lacks defined goals and success metrics. A structured plan with measurable outcomes is essential.
• Methodology: Flexibility is fine, but success must be clearly defined—what are the objectives, and how will progress be measured?
• Integration with Existing Initiatives: How will this align with Questbook and SOS? Who will manage collaboration, and do PMs/DAs have the capacity for this? We recommend waiting for SOS to help set funding priorities.
• Strategic Approach vs. Hype: Reacting late to trends isn’t ideal. If AI is a real priority, should ARDC conduct exploratory research first, or should we wait for SOS to guide the DAO’s investment strategy?

We encourage further refinement to ensure this initiative is well-structured, strategically aligned, and properly resourced.

We find the underlying premise of this proposal to be strategically compelling - positioning Arbitrum as a key infrastructure layer for AI agent coordination represents a meaningful opportunity that warrants focused attention and resources.

However, while the proposal effectively identifies the strategic importance of this initiative, in our view, its execution framework requires further refinement and solidification.

The core challenge lies in the proposal's lack of concrete implementation specifics:

• The team's operational approach and specific deliverables remain largely undefined
• The interaction model with existing Arbitrum entities (Foundation, OCL) needs clearer delineation to avoid duplicative efforts
• The absence of defined KPIs and success metrics makes it difficult to evaluate effectiveness

The agile approach to development makes sense given the rapidly evolving nature of AI technology. However, this shouldn't preclude establishing clear initial priorities and concrete success metrics.

We believe this proposal addresses a crucial strategic opportunity but would benefit significantly from a more detailed implementation roadmap, specific initial focus areas and deliverables, clear success metrics and KPIs and explicit coordination mechanisms with existing Arbitrum entities.

We would be happy to support a refined version of this proposal that addresses these elements, as ensuring Arbitrum's positioning in the AI infrastructure space represents a meaningful opportunity.

@TodayInDeFi @Atomist @chamadao @Federico @danielM @CryptoSI please read this

Although the proposal has already been paused, but since I just got to it, I also want to add a couple of my thoughts.

I often use chatgpt and the grok and other AI tools, but I almost never use AIagents, although I have been in crypto for a very long time(maybe i'm too old for this :joy:). I remember how in 2021 everyone was talking about how Metaverses are the future and everyone urgently needs to make their own metaverses. And where are all these projects now? The trend has passed and it's all over. What do I want to convey with this?

Firstly, Arbitrum has already missed this trend, so to speak, the train has left. Secondly, the trend itself has also seemingly ended. Maybe we don't need to single out a separate group for AI trends. We have wonderful programs, for example, Stylus: the project could receive a grant for development if it is useful for the ecosystem. But it seems that there are none now as far as I know(or am I wrong and we have one or a couple?), since there is no demand, the trend is fading.

But for the future, as an option, simply allocate a quota for AIagents in programs such as Stylus, for example. I think this will be a good solution, than to create a separate group and allocate a separate budget.

@TodayInDeFi @Atomist @chamadao @Federico @danielM @CryptoSI please read this

Although the proposal has already been paused, but since I just got to it, I also want to add a couple of my thoughts.

I often use chatgpt and the grok and other AI tools, but I almost never use AIagents, although I have been in crypto for a very long time(maybe i'm too old for this :joy:). I remember how in 2021 everyone was talking about how Metaverses are the future and everyone urgently needs to make their own metaverses. And where are all these projects now? The trend has passed and it's all over. What do I want to convey with this?

Firstly, Arbitrum has already missed this trend, so to speak, the train has left. Secondly, the trend itself has also seemingly ended. Maybe we don't need to single out a separate group for AI trends. We have wonderful programs, for example, Stylus: the project could receive a grant for development if it is useful for the ecosystem. But it seems that there are none now as far as I know(or am I wrong and we have one or a couple?), since there is no demand, the trend is fading.

But for the future, as an option, simply allocate a quota for AIagents in programs such as Stylus, for example. I think this will be a good solution, than to create a separate group and allocate a separate budget.

For proposals that propose such high spend, there needs to be much clearer goals and success criteria. How will we determine if this program is a success? How will the return-on-investment for such a high spending proposal be evaluated?

We think there is benefit in promoting activity like AI agents on the network, but without clear goals and success criteria it doesn't make sense to spend such large amounts.

The proposal should include at least some KPIs. Some options could be focsued on AI agent market share, AI deployments that generate revenue for the DAO through various methods, or similar KPIs that are focused on the ROI to the DAO.

Even with these additions, we think that the $500,000 budget is extremely high. There is little reason that the DAO should be adding service provider budgets that far exceed other DAO service provider budgets that clearly deliver value for their DAOs.

As it stands we would be voting AGAINST this proposal.

After a discussion with OCL, we're pausing this proposal.

OCL is already working on enabling an AI framework and has some capabilities to serve as a point fo contact (PoC) for AI builders. Our view is that this PoC service is poorly advertised across Arbitrum digital properties and we invite OCL and the Foundation to coordinate in improving access to OCL customer support services by making them more prominent and known across the DAO too. Some ideas:

• a grantee info pack or builder info pack with all the resources available could be created/mantained (similar documents already exist but for some reason are not quite meeting the mark).
• OCL and AF coordinating to organising recurring (e.g. monthly) AMAs and roundtables to map and address builder needs and signal a welcoming ecosystem.

Generally, we see similar activities have been tried already, but are in our view not enough. Our hunch is that a more relational component is needed (it's not about having the documentation, infra, and resources, but about having someone to talk to and feel hand held). We share this view simply as feedback with the caveat that we haven't researched the merits/gaps of the customer support offered (such initiative could be carried by the ARDC).

We have also realised how communication between the delegates, @Arbitrum Foundation, and @offchainlabs about ecosystem development initiatives could be improved and have proposed to the AF and OCL to organise a monthly call to share with delegates what they're working on in this areas.

cc @raam

For proposals that propose such high spend, there needs to be much clearer goals and success criteria. How will we determine if this program is a success? How will the return-on-investment for such a high spending proposal be evaluated?

We think there is benefit in promoting activity like AI agents on the network, but without clear goals and success criteria it doesn't make sense to spend such large amounts.

The proposal should include at least some KPIs. Some options could be focsued on AI agent market share, AI deployments that generate revenue for the DAO through various methods, or similar KPIs that are focused on the ROI to the DAO.

Even with these additions, we think that the $500,000 budget is extremely high. There is little reason that the DAO should be adding service provider budgets that far exceed other DAO service provider budgets that clearly deliver value for their DAOs.

As it stands we would be voting AGAINST this proposal.

After a discussion with OCL, we're pausing this proposal.

OCL is already working on enabling an AI framework and has some capabilities to serve as a point fo contact (PoC) for AI builders. Our view is that this PoC service is poorly advertised across Arbitrum digital properties and we invite OCL and the Foundation to coordinate in improving access to OCL customer support services by making them more prominent and known across the DAO too. Some ideas:

• a grantee info pack or builder info pack with all the resources available could be created/mantained (similar documents already exist but for some reason are not quite meeting the mark).
• OCL and AF coordinating to organising recurring (e.g. monthly) AMAs and roundtables to map and address builder needs and signal a welcoming ecosystem.

Generally, we see similar activities have been tried already, but are in our view not enough. Our hunch is that a more relational component is needed (it's not about having the documentation, infra, and resources, but about having someone to talk to and feel hand held). We share this view simply as feedback with the caveat that we haven't researched the merits/gaps of the customer support offered (such initiative could be carried by the ARDC).

We have also realised how communication between the delegates, @Arbitrum Foundation, and @offchainlabs about ecosystem development initiatives could be improved and have proposed to the AF and OCL to organise a monthly call to share with delegates what they're working on in this areas.

cc @raam

Thank you, @danielo, for sharing this proposal. While we support the idea of fostering on-chain AI innovation on Arbitrum, we have some concerns regarding the proposed implementation, similar to those expressed by other delegates.

The proposal lacks clarity on how the team will achieve its goals. While the context and team roles are explained, there is little detail on what the team will actually do in practice. For instance, how will DevRel identify and engage AI builders? Will they rely on hackathons, partnerships, or direct outreach? Without specifying clear strategies, there is a risk of overlapping efforts with the Arbitrum Foundation, which already has dedicated teams for ecosystem growth. Similarly, the Technical Lead’s responsibilities—such as scoping tasks and RFPs—are vague, without specifying concrete deliverables or priorities.

The absence of a defined roadmap or KPIs further complicates assessing the feasibility of this initiative. Without these critical elements, seeing how this program will deliver a meaningful impact or avoid resource duplication is difficult.

We believe this proposal could be valuable if refined to include a more structured implementation plan with clear objectives, milestones, and differentiation from existing initiatives.

I like the idea of ​​making the Arbitrum a home for AI agents.

However, as other delegates have rightly written, this proposal is too abstract. I will not repeat their arguments (I fully agree with them), but I would like to see this proposal not completely stalled due to high resistance. It seems to me that in order to concretize this proposal, we first need to formulate the tasks that the Arbitrum will perform.

Actually, the proposal should be divided into 2 parts, and you have outlined the second part (but which will require adjustment when/if we implement the first) The first part is to formulate the tasks that users need. How to do this is an open question, but I assume that there are many AI agent specialists who can formulate such tasks and themselves be participants in the second phase as coordinators or managers.

After all, we want to make what users need. And also understand what competitive advantage our product will have compared to those already on the market.

This proposal is about having support for AI builders (dev tooling, documentation, points of contact, etc.) a Hackathon is simply some prizes (so a bit of money for builders) but doesn't address the support gap. So this is complimentary to the hackathon and grant program.

The extra devs won't be hired unless the team needs them. And indeed there's a PM to coordinate with the other orgs (OCL, Questbook, Foundation, etc.). So basically we're setting up the resources to be able to do the research+coordination you suggest and then act on the findings.

The following reflects the views of L2BEAT’s governance team, composed of @krst and @Sinkas. It’s based on their combined research, fact-checking, and ideation.

While this proposal is more of an idea than a fully developed plan, we appreciate the initiative to explore how AI could be more systematically integrated into the Arbitrum ecosystem and supported by the DAO. We understand where the need for agility comes from, and even though we support creating an agile team, we need to ensure that we are not just throwing money chasing rabbits.

We do not see a plan for how this initiative will move from an idea to an executable strategy. Funding a team is a step, but the first step should be a high-level plan of what we want to achieve. Although we could discuss KPIs and details for months, this goes against the very spirit of the proposal. In our mind, a better approach to ensuring that we will not be spending money without seeing any meaningful impact is to have clear-set red flags that urge us to wind down the initiative if needed. In that way, even if we are to fund a team without any specific goal ahead of time, and even if there are no KPIs set given the ever-changing landscape of AI x Blockchain, we will have some concrete signal that things aren’t working, and we can pull the plug.

Hi @danielo, this looks like an interesting proposal but has some gaps.

While the goal behind the proposal is applaudable, we believe that the overall goal of the proposal was unclear, but we’ve seen that you mentioned the goal was for the proposed team to work on the items below.

We believe the proposal needs more alignment with AF and OCL for better clarity: e.g. how does it fit within the Trailblazer AI Grant program? What other resources are already available to support teams? Are there other plans for future grant programs?

The goal of this proposal needs to be made clearer, whether its to support and accelerate available tooling or provide devrel support for AI teams building on Arbitrum. At the moment its not very clear what the goal is but we agree its an area of opportunity for Arbitrum. @JoJo's feedback was very on point, especially on his assessment below.

Last but not least, we believe that it would be better to avoid sharing these in proposal format, and more as temperature checks to gauge interest beforehand.

Thanks a lot for your proposal Danielo.

I think it has its merit & I have seen the products in action that you and your team have built.

This being said, given some feedback is around trying to figure out whether there is some space for AI agents in Arbitrum/where this space should be: wouldn’t it make sense for the first SOS to be completed?

This would give this proposal a North Star to attach itself to.

Thank you, @danielo, for sharing this proposal. While we support the idea of fostering on-chain AI innovation on Arbitrum, we have some concerns regarding the proposed implementation, similar to those expressed by other delegates.

The proposal lacks clarity on how the team will achieve its goals. While the context and team roles are explained, there is little detail on what the team will actually do in practice. For instance, how will DevRel identify and engage AI builders? Will they rely on hackathons, partnerships, or direct outreach? Without specifying clear strategies, there is a risk of overlapping efforts with the Arbitrum Foundation, which already has dedicated teams for ecosystem growth. Similarly, the Technical Lead’s responsibilities—such as scoping tasks and RFPs—are vague, without specifying concrete deliverables or priorities.

The absence of a defined roadmap or KPIs further complicates assessing the feasibility of this initiative. Without these critical elements, seeing how this program will deliver a meaningful impact or avoid resource duplication is difficult.

We believe this proposal could be valuable if refined to include a more structured implementation plan with clear objectives, milestones, and differentiation from existing initiatives.

I like the idea of ​​making the Arbitrum a home for AI agents.

However, as other delegates have rightly written, this proposal is too abstract. I will not repeat their arguments (I fully agree with them), but I would like to see this proposal not completely stalled due to high resistance. It seems to me that in order to concretize this proposal, we first need to formulate the tasks that the Arbitrum will perform.

Actually, the proposal should be divided into 2 parts, and you have outlined the second part (but which will require adjustment when/if we implement the first) The first part is to formulate the tasks that users need. How to do this is an open question, but I assume that there are many AI agent specialists who can formulate such tasks and themselves be participants in the second phase as coordinators or managers.

After all, we want to make what users need. And also understand what competitive advantage our product will have compared to those already on the market.

This proposal is about having support for AI builders (dev tooling, documentation, points of contact, etc.) a Hackathon is simply some prizes (so a bit of money for builders) but doesn't address the support gap. So this is complimentary to the hackathon and grant program.

The extra devs won't be hired unless the team needs them. And indeed there's a PM to coordinate with the other orgs (OCL, Questbook, Foundation, etc.). So basically we're setting up the resources to be able to do the research+coordination you suggest and then act on the findings.

The following reflects the views of L2BEAT’s governance team, composed of @krst and @Sinkas. It’s based on their combined research, fact-checking, and ideation.

While this proposal is more of an idea than a fully developed plan, we appreciate the initiative to explore how AI could be more systematically integrated into the Arbitrum ecosystem and supported by the DAO. We understand where the need for agility comes from, and even though we support creating an agile team, we need to ensure that we are not just throwing money chasing rabbits.

We do not see a plan for how this initiative will move from an idea to an executable strategy. Funding a team is a step, but the first step should be a high-level plan of what we want to achieve. Although we could discuss KPIs and details for months, this goes against the very spirit of the proposal. In our mind, a better approach to ensuring that we will not be spending money without seeing any meaningful impact is to have clear-set red flags that urge us to wind down the initiative if needed. In that way, even if we are to fund a team without any specific goal ahead of time, and even if there are no KPIs set given the ever-changing landscape of AI x Blockchain, we will have some concrete signal that things aren’t working, and we can pull the plug.

Hi @danielo, this looks like an interesting proposal but has some gaps.

While the goal behind the proposal is applaudable, we believe that the overall goal of the proposal was unclear, but we’ve seen that you mentioned the goal was for the proposed team to work on the items below.

We believe the proposal needs more alignment with AF and OCL for better clarity: e.g. how does it fit within the Trailblazer AI Grant program? What other resources are already available to support teams? Are there other plans for future grant programs?

The goal of this proposal needs to be made clearer, whether its to support and accelerate available tooling or provide devrel support for AI teams building on Arbitrum. At the moment its not very clear what the goal is but we agree its an area of opportunity for Arbitrum. @JoJo's feedback was very on point, especially on his assessment below.

Last but not least, we believe that it would be better to avoid sharing these in proposal format, and more as temperature checks to gauge interest beforehand.

Thanks a lot for your proposal Danielo.

I think it has its merit & I have seen the products in action that you and your team have built.

This being said, given some feedback is around trying to figure out whether there is some space for AI agents in Arbitrum/where this space should be: wouldn’t it make sense for the first SOS to be completed?

This would give this proposal a North Star to attach itself to.

Please see here

as mentioned, this needs to be co-created. Which success criteria do you think it should have?

I’m curious to understand how this proposal fits in with existing initiatives by the Arbitrum Foundation, like the Trailblazer AI Grant Program and the broader efforts to onboard builders into the Arbitrum ecosystem.

The Trailblazer AI Grant Program is already allocating $1M to support AI-driven projects on Arbitrum. You've answered a question about differences with it saying:

However, there are already efforts in that direction as well. For example, Arbitrum is sponsoring ETHGlobal’s Agentic Ethereum hackathon, which is all about AI agents on Ethereum. This initiative seems to overlap with that—are we duplicating efforts here, or is there a clear distinction in approach?

Also, you mention:

It's unclear how you plan on these resources be used to onboard AI builders to Arbitrum. Is it via a coordinated effort with the Arbitrum Foundation and other programs like Questbook?

Finally, given the emphasis on avoiding wasted resources and “building the wrong thing,” has there been any analysis of the projects already funded by the Trailblazer program? It seems like that would provide useful insights before launching another initiative.

Hello, we appreciate the proposal.

We are not sure we understand, will the team only be responsible for providing information, or will they assist the projects in other ways?

Do you know if there are any interested candidates for the team positions you mentioned?

We see that the budget calculation is based on six months of work. Was this timeframe chosen for any specific reason?

I imagine that primarily as a point of contact. Having a human to talk to is critical to build trust. That could look like a telegram chat with the DevRel person there doing Q&A (supported by an AI to avoid repetitive questions like TogetherCrew has built for example).

There should likely be some sort of docs/wiki, and where that is built and how it looks like would be decided by the team in collaboration with AF and others. The key here is the manpower to do it. I'm specifically not suggesting we fund some documentation hub because there are already interfaces and maybe the easiest is adding a new page in the existing website or so. But all that requires a lot of user journey work and is the type of things the team would be doing anyway.

I'm thinking any team building something like this. In RnDAO we have 2-3 (Harmonica and TogetherCrew) that could have benefited form this and I'm sure there are many others. Defining the user persona is part of the work the proposed team would need to do. And as mentioned, RnDAO is not benefiting from this so we don't have the bandwidth to to this work, we're relying on delegates to help source other examples and co-build the proposal.

agreed, please provide suggestions :) as mentioned, this proposal needs to be cocreated

the program is still active as fas as I know. The difference is those grants are about funding AI builders but we're lacking the DevRel (infra, tools, documentation, customer support) component of supporting those teams. Deploying cash alone rarely works, we need to make Arbitrum hospitable to them

There’s a huge opportunity for Arbitrum here. Moving fast with an agile approach makes sense.

Potential projects I’d be really exited about:

• AI agents with the end goal of buying ARB, similar to how a16z trades memes with its own token—but instead, focusing on real investments to ultimately buy back ARB.
• Start using AI to assess proposals and run a pilot—I want to see the potential of unbiased governance.

A budget of 500K or any amount can’t be properly evaluated at this stage, as there are still many unknowns. But I’m excited about this one!