Can Industry AI Agents Compete with Claude’s Powerful Plugins?

Claude Cowork with Plugins Enables DIY Legal AI

Anthropic’s release of Claude Cowork with plugins allows professionals—lawyers, marketers, finance experts—to create AI‑driven workflows without writing code. Zack Shapiro, a Yale‑trained lawyer, used Claude’s desktop app and six custom Skills to automate the full contract lifecycle for a two‑person boutique firm, reducing operating costs by 70% after roughly 20 hours of development.

Zack’s Implementation Details

Instead of following Anthropic’s step‑by‑step guide, Zack asked Claude to analyze six months of internal conversations and identify the most impactful Skills. Claude returned concrete recommendations, leading to six core Skills:

Contract review with four distinct modes (severity rating, missing clause checklist, market clause benchmark, hand‑off to tracking Skill)

Tracking modifications directly in the XML layer of Word documents

Contract drafting with automatic template selection based on transaction type

Client communication tailored to client background

Legal research across multiple jurisdictions with cross‑validation

Policy writing that formats citations and generates regulatory submissions

These Skills encode Zack’s ten‑year practice experience into reusable decision logic, and Claude’s real‑time code generation writes the necessary XML for Word, preserving styles and cross‑references.

Limits of the Workshop Model

While the "expert + general‑AI" approach shows promise, it has clear ceilings:

What it can do: Personal workflow automation, encoding expert judgment, instant code generation for system integration.

What it cannot do: Real‑time updates across 50+ jurisdictions, support for 1,000+ concurrent lawyers, automated quality assurance beyond manual checks.

ROI: 20 hours of effort yielded a 70% cost reduction, but this relies on Zack’s elite background and willingness to assume risk; average lawyers may not see the same payoff.

Harvey’s Value Proposition

Harvey, a vertical AI product valued at $8 billion, targets the 95% of lawyers who lack Zack’s technical depth. Its core advantage is providing a pre‑built trust infrastructure—compliance certification, data‑flywheel, expert network, and auditable decision chains—so organizations can adopt AI without building custom Skills.

Organizational vs. Individual Capabilities

The article contrasts the "Zack model" (individual, high‑skill, low‑governance) with the "Harvey model" (organizational, lower‑skill, high‑governance). A table (removed for brevity) highlighted differences in responsibility, error cost, fault tolerance, and audit requirements.

Market Re‑ordering

The legal market is shifting from reputation‑based ordering to a capability‑based hierarchy. AI removes production bottlenecks, allowing senior lawyers to complete work in minutes that previously took days. This creates a "10x lawyer" effect, where judgment amplified by AI yields orders‑of‑magnitude gains in speed and quality.

Four Quadrants of Capability vs. Governance

Using a two‑dimensional matrix, the article defines:

Quadrant I (high capability + high governance): Startup opportunities that act as knowledge brokers, extracting expert judgment and curating it for mid‑level users.

Quadrant II (low capability + high governance): Harvey’s domain—large firms needing risk control, compliance, and auditable AI.

Quadrant III (low capability + low governance): Consumer‑grade tools that are cheap, simple, and sold through brand channels.

Quadrant IV (high capability + low governance): Zack’s niche—personalized, highly controlled workflows that only top experts can encode.

Future Trajectory

Two pressures will shape the evolution:

Upward compression: As base models become more reliable (hallucination < 1 %), the workshop model’s ceiling rises, but organizational adoption still hinges on trust.

Downward diffusion: If vertical AI products lower the learning curve, more lawyers will access near‑Zack capabilities, expanding their market share.

Vertical AI platforms will build strong moats through data flywheels and expert communities, while workshop‑style solutions will coexist, serving users who need maximal control.

Conclusion

Experts can fully leverage Claude by encoding experience into structured Skills, but this solves only personal efficiency. Vertical AI products like Harvey add value by offering pre‑built, auditable trust infrastructure for the majority of practitioners. AI acts as a multiplier for strong judgment, amplifying the impact of the most skilled lawyers while democratizing a portion of that power through well‑designed platforms.