Anthropic shipped spend controls for Claude Enterprise this week. If you run a consulting firm and you’ve deployed agents that touch client work, this matters more than any feature update we’ve seen in twelve months.

The problem isn’t that AI is expensive. The problem is that agentic workflows don’t stop when you hit a budget. A research agent pulling competitive intelligence for a healthcare client engagement can burn through three months of API allocation in a weekend if it decides to crawl every cited source, summarize every PDF, and build a knowledge graph of every entity mentioned. You find out Monday morning when the invoice lands.

Consulting firms operate on project economics. Every engagement has a scope, a fee, and a margin target. When an AI agent consumes $4,000 of compute on a $25,000 project, the partner conversation isn’t about innovation. It’s about whether the firm can afford to keep the tooling turned on.

Claude’s new controls let you set hard caps per workspace, per project, and per user. You can allocate $500 to the proposal agent, $1,200 to the research agent supporting a client engagement, and $300 to the knowledge agent that indexes your internal corpus. When the cap hits, the agent stops. No surprise invoices, no post-mortem budget reconciliation, no explaining to the CFO why last quarter’s AI line item doubled.

This is the governance layer that makes agentic AI viable for firms that bill by the project.

The Budget Problem Consulting Firms Hit First

Most firms that deploy agents start with a single use case. A proposal generation agent that pulls past decks and pricing into a tailored draft. A research agent that runs structured industry analysis at the start of every engagement. A knowledge agent that reads every document the firm produces and answers questions across the corpus.

The first month looks great. The agent runs, the output is useful, and the API bill is $600. Month two, someone on the team discovers the agent can pull competitive filings, synthesize case law, or build a full market landscape if you give it the right prompt. The bill jumps to $2,400. Month three, the agent is running on every active client engagement, and the invoice is $7,800.

No one did anything wrong. The agent worked exactly as designed. The firm just didn’t have a way to allocate cost to the engagement that triggered the work.

Consulting economics don’t tolerate unallocated overhead. If you can’t tie the AI cost to a specific client project, it comes out of firm margin. Do that for two quarters and the ROI conversation turns into a shut-it-down conversation.

The firms that survive this phase are the ones that build manual governance before the bill spirals. They create a spreadsheet that tracks which agent ran on which project, estimate the token cost, and allocate it back to the engagement P&L. It works, but it’s a tax on every deployment. You spend more time managing the agent budget than you save by running the agent.

Claude’s spend controls move that governance into the platform. You set the cap when you create the workspace. The agent runs until it hits the limit, then stops. The cost is predictable, the allocation is automatic, and the partner leading the engagement knows exactly what the AI tooling will cost before the work starts.

What Spend Controls Actually Let You Do

The feature set is straightforward. You can cap spend at the organization level, the workspace level, or the user level. You can set daily, weekly, or monthly limits. You can configure alerts at 50%, 75%, and 90% of the cap. When the limit hits, the agent stops making API calls until the period resets or you manually increase the allocation.

For consulting firms, the workspace-level control is the one that matters. You create a workspace for each client engagement, assign the relevant agents, and set a spend cap that fits the project budget. If the engagement is a $40,000 strategy project and you’ve allocated $1,500 for AI tooling, you set the workspace cap at $1,500. The research agent, the proposal agent, and the knowledge agent all draw from that pool. When it’s gone, they stop.

This doesn’t just prevent runaway bills. It changes how you price and scope AI-augmented work.

Before spend controls, most firms treated AI cost as a firm-level expense. You paid for the platform, you paid for the API usage, and you hoped the efficiency gain across all projects justified the cost. You couldn’t price AI into a specific engagement because you didn’t know what it would cost.

With spend controls, you can. You know the research agent will cost $800 to run a full competitive landscape. You know the proposal agent will cost $300 to generate a tailored pitch deck. You know the knowledge agent will cost $150 to index the client’s internal documents and answer questions for the duration of the project. You add those numbers to the engagement budget, set the workspace cap, and move forward.

The client doesn’t see the line item. But you know the margin impact, and you know the agent won’t blow past the allocation.

Where Consulting Firms Leak Budget Without Governance

The typical consulting firm doing $5M to $15M in revenue loses between $80,000 and $300,000 a year to work that shouldn’t require senior people. Proposal writing, research synthesis, and knowledge management are the three biggest drains.

Proposal writing is the most visible. A partner or principal spends 20 to 40 hours writing a deck for a major opportunity. They pull case studies from past projects, customize the pricing, write the executive summary, and build the project plan. Half of that work is assembly, not strategy. The firm pays $8,000 to $15,000 in partner time to produce a document that’s 60% reusable content.

Research synthesis is the hidden cost. Every engagement starts with secondary research. Industry reports, competitive filings, market data, customer interviews. A senior consultant or analyst spends two to three weeks reading, summarizing, and building a brief. The firm bills some of that time, but not all of it. And the next engagement in the same industry starts from scratch, even though 70% of the research overlaps.

Knowledge management is the long tail. Every project produces deliverables, meeting notes, and internal analysis. Almost none of it is reusable. When a new engagement needs a similar insight, the team either reinvents it or spends hours searching for the original document. The firm pays for the same thinking twice.

AI agents eliminate the repetitive work in all three categories. A proposal generation agent pulls past proposals, case studies, and pricing into a tailored draft in 20 minutes. A research agent runs structured industry and company research at the start of every engagement, with sources, summaries, and a one-page brief. A knowledge agent reads every deck, doc, and meeting transcript the firm produces and answers questions across the corpus.

But only if you can control the cost.

How Omni Ops Agents Run on Capped Budgets

We build three types of agents for consulting firms through Omni Ops: proposal generation, research, and knowledge management. All three run on Claude Enterprise with workspace-level spend controls.

The proposal generation agent starts with your firm’s past proposals, case studies, and pricing models. You give it a new opportunity brief, and it pulls the relevant content, customizes the narrative, and generates a draft deck. The agent runs in a dedicated workspace with a $300 cap. It completes the work in 15 to 25 minutes and uses $180 to $250 of the allocation. You review the draft, make edits, and send it to the client. Total partner time: two hours instead of 30.

The research agent runs at the start of every client engagement. You give it a research brief — industry, competitors, market size, key trends — and it pulls public filings, analyst reports, and news articles. It summarizes each source, builds a one-page executive brief, and flags gaps where primary research is needed. The agent runs in a project-specific workspace with a $1,200 cap. It completes the research phase in three days instead of three weeks. Total analyst time: five hours instead of 80.

The knowledge agent indexes every document your firm produces. Proposals, decks, meeting notes, deliverables, internal memos. You ask it a question — “What pricing model did we use for the last healthcare strategy project?” or “What were the key findings from the retail market analysis we did in Q2?” — and it pulls the answer with citations. The agent runs in a firm-wide workspace with a $500 monthly cap. It replaces the manual search process that consumes 10 to 15 hours of senior time every week.

All three agents stop when they hit the cap. You don’t get a surprise invoice. You don’t have to reconcile the cost after the fact. You know what the agent will cost before you turn it on.

If you want to see how these agents map to your firm’s workflow, book a 60-min Omni Audit. We’ll walk through your proposal process, your research workflow, and your knowledge management setup. You’ll leave with three things: a cost model for the agents, a 90-day deployment plan, and a list of the manual tasks you can eliminate in the first 30 days. No deck, no sales pitch, just the plan.

The Real Governance Layer Is Process, Not Platform

Spend controls solve the budget problem. They don’t solve the process problem.

The firms that get value from agentic AI are the ones that treat the agent like a team member, not a tool. You don’t hand a junior consultant a vague brief and hope they figure it out. You give them a structured task, a clear output format, and a quality bar. You review their work, give feedback, and iterate until it’s right.

The same logic applies to agents. A proposal generation agent needs a template, a content library, and a review process. A research agent needs a structured brief, a source list, and a quality check. A knowledge agent needs a document taxonomy, a citation format, and a way to flag low-confidence answers.

Most firms skip this step. They turn on the agent, watch it produce output, and assume it’s working. Three months later, they realize the agent is generating proposals that don’t match the firm’s pricing model, research briefs that miss key competitors, or knowledge answers that cite outdated documents.

The fix isn’t better AI. It’s better process.

We see this in every Omni Audit for consulting firms. The firms that get the most value from agents are the ones that already have tight processes for proposal writing, research, and knowledge management. The agent just makes those processes faster. The firms that struggle are the ones that treat the agent as a replacement for process. It isn’t.

If you’re deploying your first agent and you want a structured way to think through the process layer, we built a worksheet that walks through the setup step by step. It covers the task definition, the output format, the quality bar, and the review process. You can grab it here: Deploy Your First Business Agent. It’s a 20-minute exercise that saves you three months of trial and error.

What the Next Six Months Look Like

Anthropic’s spend controls are the first real governance feature for agentic AI. OpenAI will ship something similar in the next quarter. Google already has budget caps in Vertex AI. Every platform will have this by the end of the year.

The firms that move now have a six-month window where governance is a competitive advantage. Your competitors are either avoiding agents because they can’t control the cost, or they’re running agents without caps and dealing with surprise invoices every month. You can deploy agents with predictable cost, allocate them to client engagements, and price AI-augmented work with confidence.

That window closes when governance becomes table stakes. At that point, every firm has spend controls, and the advantage shifts to execution. The firms that win are the ones that built the process layer, trained their teams, and integrated agents into every part of the workflow.

We’re running Omni Audits for consulting firms that want to move in the next 90 days. The audit is 60 minutes. We walk through your proposal process, your research workflow, and your knowledge management setup. You leave with a cost model, a deployment plan, and a list of the manual tasks you can eliminate in the first month.

No deck, no sales pitch, just the plan. Book your Omni Audit here.

If you want to see what other firms are building with Omni, we publish case studies and deployment guides on the EDNA insights hub. Most of the content is vertical-specific, so you’re reading about firms that look like yours.

The governance problem is solved. The cost problem is solved. The question now is whether you move while it’s still an edge, or wait until it’s table stakes.