From Tokenmaxxing to Trusted Throughput — Mingsheng Hong, Ironclad
AI Engineer World's Fair 2026 · 23:04
AI contract lifecycle management
Ironclad builds contract lifecycle management software for legal, sales, procurement and operations teams. It brings drafting, negotiation, approvals, signing and post-signature analysis into one platform. Teams can create no-code approval workflows, search centralized contract data and work through integrations with Salesforce and Slack. Jurist, its AI contract partner sold alongside the platform, helps in-house counsel draft documents, assess risks and redline agreements against company playbooks, including directly in Microsoft Word.
Attorney Jason Boehmig and engineer Cai Wangwilt founded Ironclad in 2014; former DocuSign CEO Dan Springer became CEO in 2025. Its engineering connects document work with structured contract data: Smart Import, introduced in 2021, uploads, tags and indexes existing agreements for search and analysis. Jurist combines user-provided contracts, templates and playbooks with legal-specific models, multiple language models and agents to produce context-aware drafts and reviews.
The platform also supports the work that follows signature. Its procurement capabilities extract obligations such as renewal windows, rebates and payment terms, organize related supplier agreements into contract families and connect contract data to SAP procurement workflows. In February 2026, Ironclad reported more than $200 million in annual recurring revenue, with approximately one-third of new customers adopting Jurist over the preceding six months.
This archive guide covers the supplied Ironclad catalog, which currently contains one recording. It is organized by ways into the material rather than as company history. All organizational practices and recommendations below are claims presented in the recording—not assertions about Ironclad’s present-day operations, policies, or products.
In From Tokenmaxxing to Trusted Throughput — Mingsheng Hong, Ironclad, Hong argues that token consumption should not become a performance target. He describes usage dashboards as diagnostic tools for identifying adoption gaps, unusual bursts, and differences between teams, while cautioning against using them to rank or reward individuals. The recording presents “trusted throughput”—reviewed, validated, maintainable work that produces customer value—as the preferred objective.
Use the same recording for Hong’s discussion of lines of code as a traditional metric and analogy, followed by Ironclad’s reported transition from open pull requests to merged pull requests and its evolving complexity-weighted merge metric. Hong presents trusted output as a combination of objective checks, human judgment, and production or customer feedback. These are recorded descriptions and proposals; the catalog does not establish that they remain current practice.
For developer workflows, review the portions of Hong’s talk addressing what happens when AI makes code generation abundant. He argues that review and continuous integration then become bottlenecks, and that slow CI can encourage larger pull requests that are harder to review. His recommendations include an AI first pass for routine review issues, continued human accountability for architecture and security design, removal of flaky tests, and tracking ready-to-merge wall-clock time and retry counts.
For operational cost management, the recording proposes applying tighter controls after the initial adoption hurdle. Hong discusses budgets, quotas, anomaly alerts, human review, and institutional learning. Suggested token-efficiency techniques include limiting retry loops, structuring prompts for prefix caching, pruning or compacting long contexts, buying nondifferentiating infrastructure, and building internally where company-specific workflows provide differentiation. These are recommendations made in the talk, not verified statements of current product capabilities or policy.
Leaders interested in organizational adoption can follow Hong’s discussion of engineering concerns about diminished craft. He argues that leaders should engage with those concerns instead of relying solely on top-down mandates. Pair this section with the measurement discussion because the recording connects trust, incentives, accountability, and adoption.
AI Engineer World's Fair 2026 · 23:04
Affiliations reflect their AIE appearances, not necessarily current employment.