Reflections from the 2026 Accord Project Meetup

The meetup brought together the Technical Steering Committee, maintainers, contributors, and guests, with community members also joining online from around the world. It was our first in-person gathering in over a year, and it landed on a milestone worth marking: ten years since the very first commit to Concerto. This post is a look back at what we covered — the state of the community, the releases that shipped over the past year, the honest reading of where we stand, and the work we did together in the room. A separate post will follow with where we’re taking the project next.

Happy Birthday Concerto

Nearly ten years ago — November 2016 — Daniel Selman pushed a 39-line “Initial commit” to accordproject/concerto. A decade later, Concerto has grown into a mature, widely used modelling language with 133 contributors and 5,961 commits behind it. Opening the meetup by celebrating that milestone was a good reminder of how far an open, community-built standard can travel.

State of the Community

Matt opened with a review of the 2025/26 year, and it was a big one. Highlights from the past twelve months included:

• Q2 2025 — Our first live MCP↔Claude integration, and the APAP reference implementation going live at api.accordproject.org.
• Q3 2025 — All five Google Summer of Code 2025 projects delivered: a linter, a conformance suite, an AI assistant, a YAML converter, and a Playground redesign.
• Q4 2025 — Concerto v4 Alpha shipped. We also weathered an npm supply-chain attack and completed a full security remediation.
• Q1 2026 — Concerto v4 Beta, acceptance as a GSoC mentoring organisation for 2026 (which drove the largest wave of contributor PRs we’ve ever seen), and our membership application to the Linux Foundation’s Agentic AI initiative.
• Q2 2026 — The Template Stack v1.0 (Markdown Transform, Cicero, and the Archive Format), the Concerto v4 release, the Concerto Playground going live, early agentic-payments work, and a VC/DID signing proposal.

A year of step-change activity

The scale numbers told their own story:

• 1,044 pull requests opened — up 167% year over year.
• 159 new contributors joined over the year.
• A steady release cadence held throughout the surge.

We also tried to read those numbers honestly. The surge is real, but it’s GSoC-shaped: the overwhelming majority of those PRs landed between December and May, against a much lower baseline before. The right way to think about it is a funnel to convert into lasting contributors, not organic growth on its own.

Faster reviews, despite the volume

Encouragingly, maintainer responsiveness improved even as volume roughly tripled:

• Time to first response: 14.6 days → 8.3 days (−43%)
• Time to merge: 21.6 days → 16.9 days (−22%)

The GSoC engine

Google Summer of Code remains the single biggest driver of new contributors: 132 first-time contributors arrived in the Dec–Mar run-up, with 40 joining in February 2026 alone. 17.8% of contributors returned year-to-year. The base is substantially rebuilt each GSoC cycle, which makes retaining even a handful of those contributors one of our most important jobs.

The honest reading

Rather than only celebrating, we spent time naming the challenges plainly:

• Concentrated maintainership. A standard is durable only when many organisations are invested in maintaining it.
• Signal vs. maturity. GitHub stars on core repos under-signal how mature the code actually is, which can give adopters scanning for a safe dependency the wrong impression.
• Program-shaped retention. A GSoC-renewed contributor base caps how much community momentum compounds year to year.
• A closing window. As legal AI heats up, the opportunity to establish open, neutral standards is real but time-bound.

Being candid about these is what let the rest of the meetup be productive.

OpenAgreements.org

A highlight of the sessions was a presentation from Steven Obiajulu on OpenAgreements.org, an effort to publish open, standardised legal templates — drawing on sources like Common Paper, the NVCA forms, and Creative Commons — so that AI systems have a clear, compliant foundation to work from rather than free-text guesswork.

Steven demonstrated an MCP integration with Claude that discovers the right template for a request (for example, a one-way NDA), then walks through an interactive Q&A to fill it in and produce a finished document. A recurring theme was compliance: the project checks state-by-state legal changes to warn against provisions that could create real liabilities. It was a great, concrete example of the kind of ecosystem the Accord Project’s tools are built to support — and a reminder that neutrality is itself a feature that adopters and vendors value.

Web Playgrounds

We also showcased the two web Playgrounds that shipped over the year:

• The Concerto Playground lets you visualise and share domain models with non-experts, then hand off to engineers or AI for integration.
• The Template Playground lets you exercise computable contracts end to end.

Together they act as demonstrators and guided-learning tools for the Accord Project stack — a low-friction on-ramp before you drop into a full development environment.

Rolling up our sleeves

The meetup was as much a working session as a conference. After the retrospective, the group moved into hands-on building — scoping a tightly focused reference demo, pressure-testing it in a share-back at the end of day one, and refining the details of how our tooling should feel in the first sixty seconds of use. Day two opened with a GSoC 2026 mid-term review led by Diana and Sanket, before further working sessions on adopters, partnerships, and next-step ownership.

We won’t spoil the details here — the output of those sessions feeds directly into the plans we’ll share in the follow-up post.

Day two: telling the story

Where day one was about making, day two turned to telling the story — how we take the past year’s releases and momentum to a wider audience. The afternoon readout was a round-robin of the marketing and developer-experience work the group had prototyped, with owners attached. A few threads stood out.

A content and community strategy. Jonathan and Moh presented a marketing plan which includes a technical blog series, white papers, design guides, and legal-tech explainers. It also means showing up where developers and legal technologists already are: Hacker News, Reddit, LinkedIn, Discord, and YouTube. On the automation side, Tech Working Group calls are already posted to Vimeo with monthly summaries and a quarterly blog that auto-posts to LinkedIn; the plan is to extend that regular cadence to other channels and to revive the project’s dormant X/Twitter account.

Concerto, positioned for developers. Sanket previewed a comparison write-up explaining how Concerto differs from Zod and JSON Schema — built around human alignment (a Concerto model is far easier to talk through with a product owner than raw JSON Schema), richer modelling constructs (relationships, inheritance, polymorphism), and its fit for AI agents. Concerto compiles to multiple targets — TypeScript, Java classes, JSON Schema, Go — is more token-efficient than JSON, and, after a decade in the open, is well understood by today’s foundation models. The team framed this as an “olive branch” to the Zod and JSON Schema communities — even floating a Zod code-generation target — rather than a turf war.

Lowering the barrier to trying it. Several efforts converged on first-run experience. Ertugrul demonstrated publishing the project’s agent skills so they can be installed with a single npx skills add accordproject/skills, appear on the skills leaderboard, and be linked from the getting-started docs. The organisation website got a refresh that puts “try it in your browser” first, linking directly to the Concerto and Template Playgrounds. And the template library was cleaned up — templates are now categorised, and 22 clunky or duplicate ones were archived so newcomers see a curated set (they remain in the repo, just hidden by default). Monish walked through a beginner-friendly Template Stack 2.0 write-up — a blog, GitHub repo, and screenshots — that takes a reader from a plain-language prompt through to a validated, computable agreement using a set of composable agent skills.

What’s next?

This meetup was a checkpoint on ten years of work and a launchpad for the next twelve months. We came away with a clear-eyed view of our strengths and our risks, a shortlist of the releases and demos we want to get in front of people, and real momentum on turning this year’s contributor surge into durable community.

In an upcoming post, we’ll share the strategy we aligned on in Dublin and what it means for where the Accord Project is headed. In the meantime, thank you to Docusign for hosting, to everyone who travelled to Dublin or joined online, and to the whole community — ten years in, and building.