Source publication: December 29, 2025, at 10:55 a.m. UTC+8. This article draws on PANews’ published transcript of an ETHPanda Talk interview with Tomasz Stanczak, introduced as Ethereum Foundation co-executive director, Nethermind founder and Ethereum core developer. The recording date is not specified; references to forthcoming announcements and relative timelines reflect the interview, not verified developments as of publication.
Ethereum’s scaling agenda has measurable technical targets. Its coordination agenda is harder to define: how can independently operated networks, wallets and applications work together without weakening the properties that make Ethereum useful?
That distinction runs through Stanczak’s conversation with ETHPanda host Bruce. His account connects three subjects often treated separately: client development, the Ethereum Foundation’s role in an independent ecosystem, and emerging infrastructure for AI agents. Across all three, the central question is what shared infrastructure can enable—and which trust assumptions remain unresolved.
From client implementation to infrastructure business
Stanczak described founding Nethermind in 2017 after working in financial technology, including at Citi in London and a hedge fund. His initial ambition was to build software for his own on-chain transactions. Reading Ethereum’s Yellow Paper led him into implementing the Ethereum Virtual Machine and, eventually, a complete client.
According to his recollection, a usable client emerged after roughly a year. He initially financed the company himself, with Greg joining as an early engineer and co-founder. An Ethereum Foundation grant of $50,000, received approximately one to one-and-a-half years after the company began, helped mark the transition to formal hiring.
Two milestones illustrate the difference between technical delivery and commercial validation. The client’s first successful synchronization with Ethereum mainnet demonstrated that the implementation worked. The xDAI network becoming Nethermind’s first paying customer, requesting customized extensions, supplied an initial business use case.
Stanczak said the original commercial thesis included financial institutions operating clients locally for data extraction, remote procedure calls and transaction submission. He reported that Nethermind eventually served those customers, despite early skepticism about whether client development could support a business.
He also estimated that Nethermind software ran on approximately 25%–30% of Ethereum mainnet nodes. The transcript supplies neither a measurement date nor a dataset or counting methodology, so that figure should be read as his interview estimate, not an independently established client-distribution measurement.
Scaling has targets; user experience lacks a shared definition
The interview framed the Foundation’s priorities around scaling L1, expanding blob capacity for L2s, and improving interoperability and user experience.
Stanczak considered the last category the broadest and hardest to evaluate. L1 scaling can be expressed through numerical objectives: he cited a gas-limit target of 100 million per block, followed by 300 million. Elsewhere in the discussion, he described reaching 300 million as an unfinished two-year ambition. These were goals in his account, not evidence that the network had already reached them.
User experience does not have an equivalent single measure. It can mean lower latency, stronger privacy, safer asset transfers, better wallet interfaces or more suitable infrastructure for institutions. Different stakeholders may therefore agree that UX needs improvement while disagreeing about what successful delivery means.
Blob scaling presents a related but distinct problem. Greater data capacity must be delivered without sacrificing adequate decentralization and security. Stanczak described the roadmap as comparatively concrete, while acknowledging that implementation choices and longer-term data-availability design still required work.
The research implication is that throughput and coordination need different evaluation criteria. A capacity target can be tracked numerically; a claim that Ethereum now feels like one integrated system requires examination of the actual user journey and its security assumptions.
L2 interoperability is a trust problem, not just an interface problem
Stanczak identified bridge security, permissionless access and preservation of L1 properties—including censorship resistance—as core interoperability challenges.
EVM compatibility may make networks easier to connect, but it does not by itself establish that assets can move between them safely or without reliance on intermediaries. He cautioned that many existing bridges and solutions associated with the Open Intents Framework, or OIF, did not fully satisfy those requirements. Better interfaces can also conceal risks rather than remove them.
He discussed the Ethereum Interoperability Layer, or EIL, as an effort originating with the account-abstraction team. Its intended direction, in his account, included permissionless interoperability without third-party intervention. The interview described that as a design objective and referred to a planned Devconnect announcement; it did not establish that the objective had already been achieved.
Other barriers are less dramatic but still consequential:
- Consistent address representation across networks.
- Compatibility of tools and applications, including multisignature functions.
- Clear representation of stablecoins and other assets across chains.
- Wallet interfaces that help users recognize the same asset on different networks.
Stanczak pointed to recommendations within Kohaku Wallet and its SDK as part of this work. The intended experience was cross-L2 activity that feels as straightforward as using a single chain. That intended simplicity should not be confused with identical underlying trust assumptions.
Commercial coordination remains a separate constraint
The obstacles are not exclusively technical. Stanczak distinguished engineers’ willingness to solve shared problems from operators’ concerns about branding, positioning and competition.
He described a proposal to bring L2 engineers and researchers together for dedicated meetings in the following year. The proposal followed feedback that L2 core developers lacked the regular in-person contact familiar to L1 client teams.
His broader point was that interoperability begins with teams choosing to integrate. Technical standards alone cannot ensure that commercial decision-makers will support that integration.
The Foundation as coordinator rather than command center
Stanczak described the Foundation’s role as gathering requirements from participants ranging from cypherpunks to institutions, communicating those requirements to core developers, and supporting focused research without dictating the outcome.
He said he and Hsiao-Wei Wang generally divided executive responsibilities by allowing whoever first took ownership of an issue to handle it. His work leaned toward external communication, while Wang handled substantial internal operations, treasury and DeFi-related responsibilities. Important initiatives could still be shared.
For future priorities, he discussed finality, continued scaling, privacy and security. These remained under discussion. He reported demand from institutions, L2 teams and users for shorter block times and faster finality, and described a separate 50-person privacy cluster with its own planning work.
Community support was also changing in his account. Rather than relying mainly on unconditional financial grants, the Foundation was emphasizing introductions, visibility, education, local hubs and connections to sponsors. He mentioned collaboration with Hong Kong Polytechnic University and plans for a Hong Kong community hub.
The stated approach was to strengthen independent builders rather than make the Foundation the operator of every initiative.
AI agents: payments, identity and verification
Stanczak’s argument for Ethereum’s relevance to AI centered on digitally native agents that need to pay, identify counterparties and evaluate their work.
He distinguished two pieces of infrastructure:
- x402: described as supporting payment requests through HTTP, with verification on a blockchain, and not limited to AI agents.
- ERC-8004: described as agent-oriented infrastructure for identity, registration, service discovery, reputation and verification of claims or work.
In his explanation, ERC-8004 could support agents acting under delegated user authority as well as autonomous agents associated with on-chain accounts. Possible verification arrangements included external on-chain and off-chain validators re-executing portions of a process to check results.
These mechanisms address different questions. An identity declaration identifies a participant; reputation provides information about it; validation checks a claim or result. None should be treated, on the strength of the interview alone, as a universal guarantee that an agent is trustworthy.
Stanczak said early dashboards, implementations and agents were being built around ERC-8004. He also described a decentralized-AI team led by David Crapis, formed roughly two months before the interview, and reported coordination involving Google, Cloudflare and Coinbase around agent payments. Those statements describe his account of ecosystem activity, not independent adoption measurements.
He expected payments to be a stronger near-term driver than some forms of decentralized training or verification, which he acknowledged might develop more slowly or remain niche.
Agent-assisted governance remains experimental
The discussion of ChaosChain moved from payment infrastructure to the influence AI might have on protocol decisions.
Stanczak suggested that developers already consulting language models could gradually rely on them more heavily. His references to AI influencing 5% and later 10% of thinking were illustrative, not measured findings.
The deeper concern was proposal generation and filtering. If AI makes it inexpensive to produce many technical ideas, governance may depend increasingly on who can validate them—and who controls which proposals become visible. Selectively publishing favorable proposals could influence social consensus even if humans retain formal decision-making authority.
He described ChaosChain-related work splitting into directions involving ERC-8004-based agent applications and attribution of collaborative agent work for rewards. Nethermind, meanwhile, continued exploring AI assistance for DAO decisions.
His possible progression—from agents commenting on research and EIPs to participating in AllCoreDevs discussions—was a prospective path, not a report of completed deployment.
The interview’s strongest common thread is therefore a boundary: shared infrastructure can make coordination easier, but it does not eliminate the need to examine authority, incentives and verification. That applies to bridges between L2s as much as it does to AI agents proposing changes to Ethereum itself.