Clio Duo vs Smokeball Archie vs a Custom AI Agent

By Jude Lee · · Comparison

Attorneys and a paralegal reviewing AI-assisted matter documents on a laptop in a small law firm conference room

The comparison most firms are actually making

When a managing partner searches for “law firm AI software,” the honest question underneath is narrower: my practice management vendor keeps emailing me about their AI assistant — is that enough, or do I need to build something?

Those two options aren’t really competing. They sit at different layers. Built-in AI assistants — the ones Clio, Smokeball, MyCase and document platforms like NetDocuments have shipped or announced under various product names — operate on the data inside their own product, with a fixed set of capabilities the vendor decided to ship. A custom AI agent is a system you configure: it reads and writes across multiple tools, follows instructions you wrote, and takes multi-step action.

The legal tech announcement cycle blurs this constantly. “Agentic” is now a marketing label applied to shipped software as readily as it describes something you build yourself. Document management and e-discovery vendors have both moved in that direction. Read the documentation and ask what the tool can actually do on your data — not what the press release calls it.

What built-in practice management AI is genuinely good at

The strength of an assistant living inside your practice management system is context and permissions. It already knows which matters exist, who the client is, what’s billable, and who is allowed to see what — because it inherits the platform’s own access controls. That’s not a small thing. Half the work of any custom AI project is exactly that plumbing.

In practice, this layer tends to do well at: summarizing a matter or a long document already stored in the system; drafting a first-pass letter or email from matter data; surfacing time entries you forgot to capture; searching across matters in plain language; and light document generation. If your firm’s pain is “nobody can find anything” or “we lose billable time,” the built-in tools plus disciplined process may close most of the gap. Our time tracking and billing automation guide covers where that specific leak actually comes from.

What this layer is generally not built for: taking a sequence of actions across systems the vendor doesn’t control, applying your firm’s idiosyncratic review standard the same way every time, or handling a workflow that starts in a PDF pile and ends in three different places.

What a custom agent adds — and what it costs you in ownership

A custom agent is an AI assistant (Claude, or a comparable model) connected to your firm’s tools through MCP — the Model Context Protocol, an open standard for giving an AI governed access to specific data and actions. Instead of a chat window that answers questions, you get something that can be handed a multi-step job: read the new intake submission, check the parties against the conflicts database, draft the engagement letter from the right template, and queue it for a human to approve and send.

Two building blocks make that repeatable. First, the connection itself — see our walkthrough on connecting Claude to Clio via MCP and, when your data lives in places no vendor has an integration for, building a custom MCP server over your firm’s matter data. Second, skills: packaged instructions that teach the assistant to perform one job your firm’s way, every time — the same reason you have a form file.

The cost isn’t only the build. It’s ownership: someone has to test it, update it when a vendor changes an API, and notice when output quality drifts. A vendor absorbs that for you. That trade is the whole decision.

Built-in AI in your practice platform
  • Included or lightly priced on top of software you already run — if your plan tier covers it.
  • Inherits platform permissions and audit trails; no separate access model to design.
  • Zero build time and no internal owner required.
  • Scoped to that vendor’s data and roadmap; behavior changes when the vendor decides it does.
  • Best for summarization, search, drafting, and billing hygiene inside one system.
Custom agent over MCP
  • You define the steps, the guardrails, and the output format.
  • Spans practice management, email, e-signature, document storage, and spreadsheets.
  • Requires build effort, a named internal owner, and ongoing maintenance.
  • Can break quietly when a vendor changes or deprecates an API — often noticed only when output goes wrong, so you need monitoring you built yourself.
  • Best for cross-system, high-volume, rules-heavy workflows where consistency matters more than flexibility.

A rough test for which layer a job belongs to

Run each candidate workflow through four questions:

  1. Does it live in one system?

    If the input and output both sit inside your practice management platform, try the built-in AI first. Building a custom agent to do what you already pay for is an easy way to waste an AI budget.

  2. Does it require taking actions, not just producing text?

    Creating a matter, sending for signature, updating a field, filing to the right folder — actions across tools are where agents earn their keep. Pure drafting usually isn’t.

  3. Does the output have to look identical every time?

    A standardized review memo, a lease abstract with fixed fields, a chronology in your format. Repeatability is a skills problem, and skills are something you author. See our approach to repeatable document review skills.

  4. Does it happen often enough to matter?

    Volume × minutes saved × how much you’d pay to buy that time back. If the workflow runs twice a month, automate it with a checklist and move on.

Treat this as a heuristic, not a score. Our rule of thumb: the more of these a workflow answers yes to, the better a custom build looks — and a workflow that spans systems and needs identical output and runs constantly is the clearest candidate, because that combination is exactly what no single vendor’s tool covers. Adjust the cut line to your firm: a two-lawyer practice with no technical owner should demand more yeses than a firm with an operations manager who already maintains integrations.

When the answer is one or two yeses, the middle ground is usually right and usually skipped. That middle ground has real names: vendor-native automation (document automation and workflow templates inside Clio or Smokeball, Outlook rules, DocuSign or Lawmatics templates) and general-purpose iPaaS/no-code tools (Zapier, Make, Microsoft Power Automate) that move data between systems on triggers without any model in the loop. These are cheaper to build, far easier to debug, and they don’t hallucinate. Our no-code legal automation guide goes deeper.

The failure mode isn’t buying the wrong AI. It’s building a custom agent for a workflow your existing software would have handled if anyone had read the release notes.

Model the money yourself — don’t accept a vendor’s number

Nobody can tell you what AI will save your firm, and any article quoting a precise hours-per-week figure is guessing. Build the estimate from your own data instead:

tasks/month × minutes saved × 12
Annual hours recovered
Requires your own task counts from a 30-day sample
recovered hours × realistic billable rate
Upper bound on value, only if those hours get billed
Requires your firm's actual realization rate, not standard rate
license + build + annual maintenance
Total cost side
Requires a real quote plus an estimate of the internal owner's hours

The discipline that matters: recovered hours only convert to revenue if they’re reallocated to billable or business-development work. Time saved that becomes time browsing is a real quality-of-life win and a zero-dollar financial one. Be honest about which you’re buying. For a fuller version of this calculation, see our law firm automation ROI walkthrough.

The disadvantages nobody’s demo covers

Generative AI fabricates confidently, including citations — a failure mode courts have sanctioned lawyers over. Every output that touches a filing, a client deliverable, or a legal conclusion needs a human who verified it against the source. That’s not a temporary limitation to be engineered away; it’s the operating assumption.

On ethics, the American Bar Association’s Formal Opinion 512 (July 2024) addresses generative AI and lawyers’ duties around competence, confidentiality, communication, supervision, and fees. State-level guidance exists too: the State Bar of California published practical guidance on generative AI for lawyers, and the Florida Bar issued Ethics Opinion 24-1 on the subject. Rules vary by jurisdiction — confirm what binds you with your own state bar’s ethics resources or a qualified professional before routing client confidences through any tool, built-in or custom. The vendor’s security page is not the analysis.

And the quieter risk: if your firm doesn’t provide a governed option, people use ungoverned ones. Our piece on replacing shadow AI rather than banning it makes that case.

Where is your firm losing billable hours?

Get a free automation audit: we map your intake-to-invoice workflow and show you exactly what's worth automating — before you spend a dollar.

Get a free automation audit

Related guides