The default stack for an in-house Legal Ops team that has cleared the Stage-2-to-3 transition on the Legal Ops maturity model — past the spreadsheet-and-shared-drive era, into the operationally-disciplined era where contract velocity, outside-counsel governance, and AI adoption all matter at the same time.
One constraint dominates every decision below: Harvey sells in blocks of roughly 20-25 seats on a 12-month commitment. There is no five-seat senior-attorney pilot. For a 20-attorney department that means Harvey is not a layer you add to a few people — it is a second purchase larger than everything else in the stack combined. The cost section treats the stack as two budgets with a cliff between them, because that is how it actually buys.
How the pieces fit
- Ironclad is the CLM backbone. Every contract enters via Ironclad intake, gets routed by playbook, signed via Ironclad signing, and lives in the Ironclad repository post-signature. The contract review SOP is encoded in Ironclad workflow rules, so the routing logic is the system of record — not a wiki page a paralegal has to remember. Its 2026-08-05 procurement release added AI Obligation Extraction (renewal windows, discounts, termination rights and payment terms as tracked fields) and a Contract Family Agent for supplier-contract hierarchies, which is what makes it carry buy-side paper rather than only the agreements you send out.
- Spellbook handles the day-to-day drafting and redlining. Spellbook Suite runs review, drafting, comparison and playbook checks in Microsoft Word, where the lawyers already are; Associate is its agent for matters that span several documents at once. It now reaches past Word into email, Slack and Salesforce. This is the layer that touches the most matters, because most in-house work is commercial-agreement redlining, not litigation.
- Harvey is the reserved high-bar layer for matters that exceed per-seat drafting AI. Its Workflow Agents run pre-built, no-code sequences over a document set — M&A diligence, credit agreements, lease portfolios — breaking the task into steps and pausing for attorney review. Since the June 2025 LexisNexis alliance it also answers primary-law questions in place: Ask LexisNexis has been in open beta for all Harvey users since August 2025, with Shepard’s citation checks and LexisNexis’s Protégé assistant inside the Harvey workspace. Harvey owns no corpus of its own — it rents that authority from a company that also sells research directly.
- Thomson Reuters CoCounsel is the research backbone. CoCounsel Legal became agentic on 2026-08-20, generally available in the US and built on Anthropic’s Claude Agent SDK: Westlaw Brief Builder takes a litigator from research to a first-draft brief in one workflow, Workspaces hold a matter’s documents and precedent together, a drafting agent takes natural-language instructions inside Word, and Tabular Analysis reviews up to 10,000 documents against up to 100 questions per run. Deep Research Verify checks whether the cited authority actually supports the assertion — the check that matters when the answer has to survive the GC or a regulator, not merely read well.
- Claude is the general-purpose AI surface for the long tail — matter summarization, outside-counsel invoice review, vendor diligence questionnaire response, matter status digests for the GC — and, since May 2026, the front door to the research layer as well. It is the horizontal AI that absorbs work with no dedicated tool, and the one seat every legal-ops person gets, not just the attorneys.
The handoffs that make it a stack
The value is in the seams, not the individual tools:
- Draft in Spellbook → route through Ironclad. A commercial agreement is redlined against playbook in Word, then submitted to Ironclad intake, where it inherits the approval and signature workflow. Nobody re-keys terms between the two.
- Ironclad repository → Claude for reporting. Signed contracts and their obligation fields feed Claude-generated matter digests and status rollups, so the weekly report to the GC is a query rather than a manual chase. Ironclad ships an official MCP server — remote, OAuth, and today supporting Claude, ChatGPT and Slackbot clients — so this is a live connection, not a scheduled export.
- Claude → CoCounsel Legal over MCP. Announced 2026-05-12 and expanded that August, this puts 1.9 billion Westlaw and Practical Law documents and 1.4 billion KeyCite validity signals behind a Claude prompt, with a citation ledger that traces every source in one click. A paralegal drafting in Claude can pull grounded authority without opening a second product. This handoff is what makes the cheapest seat in the stack worth more than its price.
- Escalate from Spellbook → Harvey. When a matter outgrows per-seat drafting AI — an acquisition, a novel financing — the same document set moves to a Harvey Workflow Agent for structured extraction and multi-step review. Spellbook handles volume; Harvey handles depth.
Why this combination
The mid-market in-house team needs three things at once: workflow infrastructure (Ironclad), specialist legal AI (Spellbook, plus Harvey when depth justifies it), and a grounded research layer (CoCounsel) reachable from a general-purpose assistant (Claude). Most teams skip one of those layers — buy specialist AI without the CLM, or buy the CLM without committing to AI — and end up with a stack that compounds process debt rather than cutting legal cycle time.
Two overlaps are worth naming rather than discovering after signature. Ironclad Jurist reached general availability inside Microsoft Word in June 2026, generating tracked changes and playbook rationale on third-party paper, not only on contracts drafted in Ironclad. That is the same job Spellbook does. Buying both is defensible — Spellbook is the deeper drafting product and Associate has no Ironclad equivalent — but run them side by side on your own paper for a month first, because for a team with straightforward commercial agreements Jurist may be enough on its own. Likewise, Harvey and CoCounsel now overlap on research, which they did not in 2024. The axis is no longer capability; it is who owns the corpus and who you already pay.
What it costs — two budgets, not one
For a 20-attorney team:
Without Harvey: roughly $130K-$260K/year.
- Ironclad — quote-only; the pricing page publishes no figures and prices on products, users, workflows and deployment path. Vendr tracks a median ACV of $40,000/year across 363 purchases, range $15,000-$104,272, average savings 20.6%. Budget the median, not the top of the range, unless your contract volume is genuinely enterprise.
- Spellbook — quote-only since mid-2026. The Solo $99 and Team $179 per-seat tiers it used to publish are gone from the site; pricing is now set by license headcount, with a 7-day trial as the only self-serve door. Third-party trackers put the entry seat anywhere from about $20 to $99 and enterprise from about $199 to $350 per seat/month — a spread wide enough to be evidence that nobody outside Spellbook’s sales team knows. The one usable signal is that the reported enterprise seat sits near double the $179 Spellbook last published. Plan on $50K-$85K/year for 20 seats and get the quote before you commit the line.
- Thomson Reuters CoCounsel — quote-only, sold per attorney inside broader Westlaw and Practical Law packaging. 2026 third-party reporting puts the bundled rate at $300-$600/user/month, so $72K-$144K/year at 20 attorneys. Standalone CoCounsel pricing is rare; if you already buy Westlaw, this is an uplift on an existing line rather than a new one.
- Claude — $20/seat/month billed annually, $25 monthly, on Team plans covering 2 to 150 people. Roughly $6K/year for 25 seats including non-attorney legal-ops staff. A rounding error against the rest, and the highest-return seat in the stack now that it reaches CoCounsel directly.
Adding Harvey: a further $240K-$600K/year, and it is not divisible. Reported mid-market rates run $1,000-$2,000 per seat/month against a 20-25 seat minimum and a 12-month term, and the LexisNexis integration is billed separately at a reported $400-$600 per lawyer/month on top. Community reports of 40-60% reductions against a written competing offer are consistent enough to make engineering one worth the effort. Renewal uplifts of 10-25% are common where the contract has no cap — negotiate the cap in the first term, not the second.
The practical consequence: Harvey roughly triples the cost of this stack, so it is a platform decision, not a line item. A department that cannot defend a quarter-million-dollar annual commitment on its deal profile should run the four-tool stack and send genuinely exceptional matters to outside counsel, which is what the money would otherwise be buying anyway.
Common variations
- Earlier-stage in-house team. Drop Harvey and CoCounsel; replace Ironclad with Concord or Juro. Spellbook plus Claude plus a lightweight CLM serves a 3-10 attorney team well. Swap rule: make this move when contract volume is under roughly 500/year and there is no live litigation exposure.
- Harvey rival on the high-bar layer. Swap Harvey for Legora — its chief rival after a 2026 mega-raise — when your team wants tighter in-document collaboration and European-origin drafting, or when Harvey’s seat minimum and procurement cycle do not pencil out. Given the size of the Harvey commitment, run both in a paid pilot on your own matters before signing either.
- Consolidate the Word layer. Drop Spellbook and run Ironclad Jurist for in-Word redlining instead. Swap rule: take this when your paper is mostly standard commercial agreements, your playbook is already encoded in Ironclad, and nobody needs Associate’s multi-document work. It removes a vendor and a seat line at the cost of drafting depth.
- Heavy litigation profile. Add Relativity or Everlaw for eDiscovery, and a matter-management platform (Onit or similar) on top of Ironclad. Swap rule: trigger this when you have more than a handful of active litigation matters at any given time.
- Heavy CLM customization needs. Replace Ironclad with Agiloft for organizations whose contract workflow is too non-standard for off-the-shelf CLM.
- Budget-constrained. Spellbook plus Claude plus Concord at roughly one-fifth the cost of the four-tool build; gives up the AmLaw-grade research layer but covers most of the value.
When this stack is the right pick — and when it isn’t
Pick this stack when you have 10-50 attorneys in-house, mid-market contract volume, litigation that overflows to outside counsel rather than running in-house, and a mandate to raise throughput per attorney without adding headcount. It assumes you can run a real procurement cycle against three or four quote-only vendors — if that is not true this quarter, start with the budget-constrained variation and grow into the rest.
Skip it when you are under 10 attorneys with low contract volume (the budget-constrained variation is the honest answer), when outside-counsel spend dominates your legal budget (invest in legal spend management e-billing first — the stack does not address that), or when your work is litigation-heavy rather than transactional (the eDiscovery variation, not this default, is your starting point). If a single specialist AI would cover 80% of your pain, buy that one tool and revisit the full stack when the second and third pain points show up.
What this stack does NOT replace
- Outside counsel for specialist regulatory work, securities matters, and material litigation
- A privacy/DPO function for GDPR and EU AI Act compliance
- A dedicated legal spend management e-billing system if outside-counsel spend exceeds roughly $5M/year (Brightflag, Onit, BusyLamp)
- Court-filing systems for active litigation