The agent layer. How AI plugs into delivery.
AI in consulting is not the copilot in the sidebar. It is the layer underneath delivery, trained on the firm's IP, governed by the firm's guardrails.
Most firms have done AI wrong, twice. They pasted client data into a public chat window. Then they bought a copilot licence. Neither changed the delivery economics.
The agent layer is something different. It sits inside the firm's system. It is trained on the firm's IP. It runs the first 60 percent of every deliverable.
What the agent layer is.
Not a chatbot. Not a copilot. Not a productivity tax.
The agent layer is a set of named, governed agents that map one to one with the firm's delivery method. Discovery agent. Proposal agent. Estimating agent. Status report agent. Risk register agent. QA review agent. Each one has a job. Each one runs in the firm's voice. Each one is auditable.
The agent never replaces the consultant. The agent replaces the gruntwork that was stealing the consultant's time.
How agents plug into delivery.
A new engagement starts. The discovery agent runs the structured interview using the firm's question set. The estimating agent sizes the work from comparable engagements. The proposal agent drafts the SOW in the firm's voice. A senior partner reviews, shapes, signs.
The engagement begins. The status report agent writes the weekly client update from the live data. The risk register agent surfaces what to watch. The QA review agent catches the things that would have come back as red ink from the client.
At each step, the agent is a named participant in the workflow. Not an aside.
How they get trained.
On your IP, not someone else's. The agent inherits the playbook from the Watch Me Do It interviews. The prompts are versioned. The outputs are reviewed. When the firm's method evolves, the agents evolve with it.
This is the part most platforms get wrong. A generic AI tool is generic. It cannot beat the in house team because it does not know what the in house team knows. An agent trained on your IP does.
Governance.
Every agent run is logged. Every output is reviewed. Every prompt change is tracked. The audit log answers the question regulators and clients will eventually ask. Who wrote this. What inputs did it use. What was the human review. What changed in the prompt over time.
Operator grade. Not science project.
The hard part.
The hard part is not the technology. The hard part is encoding the method properly. If the IP is fuzzy, the agent is fuzzy. If the IP is sharp, the agent is sharp.
That is why we do the IP audit and the Watch Me Do It interviews before we touch any prompts. Most firms want to start with the agent. We start with the operator.
Once the method is encoded properly, the agent layer compounds. Every engagement makes the agents sharper. Every revision improves the next engagement. The firm gets better while the team gets less tired.
Apply for the founding cohort if you want to see the agent layer running on your own IP.
Get the next note when it ships.
One short piece every two weeks. Consulting operating model, the agent layer, margin discipline. Operator length. No tracking pixels. Unsubscribe in one click.
See if your firm fits the wave.
ConsultancyOS opens public availability after Wave 3. Founding cohort firms get founder led onboarding and pricing locked for life. Apply and we will tell you which wave you fit.