X07 roadmap (next 12 months)

Goal

Make X07 the deterministic, certifiable execution substrate for agent-written software: code that agents (or transpilers) produce runs sandboxed, budgeted, replayable, and provable. Direct authoring by agents and humans stays a supported surface, with further language investment gated on measured evidence (see the eval gate below).

Positioning (updated 2026-06)

The trust half of X07 — deterministic worlds, budgets, record/replay, structured diagnostics with quickfix coverage, XTAL, certification — grows more valuable as code generation gets cheaper, because verification and review become the bottleneck. The language-surface half is the adoption barrier (agent case studies and the 2026-06 strategic review agree). The roadmap therefore leads with the substrate, and treats language-surface work as gated, evidence-driven investments.

The eval gate

labs/agent-eval/ holds a comparative benchmark (agents solving identical tasks in X07 vs Python/Rust) with a pilot result and a scaled runbook with a predeclared decision rule. The scaled run decides:

• pass → prioritize RFC 0002 (expressiveness floor: records, enums + match, string, f64) and re-run the eval after.
• fail → park direct-authoring guidance; X07 proceeds substrate-first (transpile target + verification + sandboxing), and language-surface work stops after the x07text projection.

Track 1 - Language and toolchain

• agent DX: did-you-mean suggestions on unknown symbols, fuzzy x07 doc lookup, behavioral summaries for stdlib exports, structured diagnostics in x07 run failure reports (landed 2026-06)
• x07text projection: lossless text surface via x07 ast to-text / from-text (RFC 0001, landed 2026-06); extend to fmt/docs integration
• run the scaled comparative eval (the gate above)
• RFC 0002 expressiveness floor: design now, implement only if the gate passes
• stabilize the core x07AST schema surface; tighten compatibility guarantees
• expand verification and trust tooling

Track 2 - MCP and ecosystem interoperability

• harden x07-mcp for wider external use
• surface the new doc summaries and x07text through MCP tools
• produce end-to-end demos showing X07, MCP, and trust/review artifacts together

Track 3 - Ecosystem scope (narrowed 2026-06)

Active: x07, x07-mcp, x07-registry, x07-wasm-backend, hardproof.

Maintenance mode until the eval gate passes and at least one external adopter exists: x07-studio, x07-forge, x07-crewops, x07-tactics, x07-device-host, x07-web-ui, x07-registry-web, platform repos. These receive security and compatibility fixes only. Rationale: they duplicate mature mainstream ecosystems, multiply single-maintainer load, and none of them move the core bet.

Track 4 - Adoption and governance

• one design-partner engagement with a team that runs untrusted agent-generated code (agent platforms, sandbox providers) — a single real external user outranks additional maintainers
• publish the scaled eval results, whatever they say
• move from founder-led governance toward multi-maintainer governance as external contributors appear

Targets

Q3 2026

• scaled comparative eval executed and published
• x07text surfaced in agent quickstart + MCP tooling
• first design-partner conversation underway
• stronger release and backport policy

Q4 2026

• eval-gate decision recorded (RFC 0002 go / substrate-only)
• if go: records + match implementation begins behind a schema bump
• external adopter case study or an honest writeup of why not yet

Growth metrics

• comparative eval pass-rate gap vs Python (the core number)
• external production or pilot adopters
• external contributors with merged PRs
• published roadmap updates each quarter