AgentMesh

Part 3 of 3 · Part 1: The anatomy of a mesh · Part 2: The reward is the business goal

Processes are how a mesh reaches its goals

AgentMesh published this article in November 2026.

Most of how a company works lives in people's heads: who does each step, where a person signs off, what has to be recorded. Generative AI now makes it fast to write that knowledge down as processes and to change them as the business learns. AgentMesh offers 170 starter processes to begin from. A mesh runs those processes with the agents on its roster, inside the process's gates and the mesh's rules, and improves both: its agents get better at their steps, and its processes get more efficient.

A goal says what the business wants, and a process says how one piece of work gets done.

The marketing mesh from Parts 1 and 2 has one goal that matters most: qualified pipeline up 20% this quarter, at $180 or less per lead. No single piece of work delivers that. Paid search, nurture sequences, webinars and account-based plays all feed it, and each of those is a process: a piece of work the company runs again and again, the same way each time.

A process does not know why it runs. It knows its own steps and its own target, such as a cost per result for one paid search campaign. The goal knows what the business wants and nothing about how the work is done. A mesh holds both, so it can run the processes the goal calls for and judge each one by what it does for the goal.

A written process says who does what, where a person signs and what each run records.

The AgentMesh Catalog holds 170 starter processes across five departments, each written down step by step. Each one names when a run starts and ends, every step and who does it, what each step hands to the next, where a person must sign, and what every run must record. The parts that belong to one company, such as the owner, the spending caps and the targets, are left blank for that company to fill in. A company can start from one of them, or have an AI assistant write down a process its team runs today from what the people doing it describe.

Run Paid Search Campaign is a typical one. A run starts when someone commits a budget and says what it must return, and it ends when the buy is closed and the spend is accounted for. In between are 13 steps:

  • The campaign manager writes down what one result is worth.
  • The search optimizer researches demand, and the media buyer builds the keyword plan.
  • The ads are written, the landing pages are readied, and conversion tracking is tested with one real submission.
  • Brand and legal sign the ads and pages as named people, a step the process marks as never automated.
  • The media buyer launches and tunes bids, and every bid move is recorded with its reason. A move above a set size waits for a person.
  • The spend is reconciled against the invoice, cost per result is reported against what the budget was meant to buy, and the team records what it learned.

That is most of what a mesh needs to run a piece of work safely. What the process does not say is which paid search campaigns to run, how much of the quarter's money each should get, or whether paid search is the best use of that money at all. Those are questions about the goal.

Each part of a process lands on a part of the mesh.

A process is written once and run many times. Each time it runs, the mesh supplies what the document leaves open, and the document supplies what the mesh must enforce.

The part of the processThe part of the mesh it lands onIn the paid search process
RolesThe roster: each role is filled by an agent or a person, from any vendorThe media buyer is the agency's agent, analytics is the ad platform's, and legal is a person
Hand-offsHand-offs between agents, across harnessesThe keyword groups go from the media buyer to the campaign manager, who writes the ads
Human gatesApprovals that wait for the named personBrand and legal sign the ads and pages, never an agent
Blanks the company fills inRules and budgets that bind every agent in the runThe daily spend cap, and the size of bid move that needs a person
What every run must recordThe ledger, one signed entry per stepEvery bid and budget move, with its number and reason
The process's own targetsGoal telemetry, read as early signals for the business goalCost per result, and days from budget to live ads

The last row ties processes to the scorecard in Part 2. A process's targets are the scorecard's internal process measures, and they are the early signals that show whether pipeline will follow.

The mesh manager decides which processes to run and how much each one gets.

A quarter's pipeline goal is served by several processes at once. The mesh manager decides which ones to run, how often and with how much of the budget. It starts paid search campaigns, builds nurture sequences, books webinars and opens account-based plays, then reads each process's own target as results come in.

The mesh manager runs the processes the goal calls for Qualified pipeline up 20% this quarter at $180 or less per lead Mesh manager Picks the processes Splits the budget Reads each target Run Paid Search Campaign Reports cost per result Build Nurture Sequence Measured once it is live Produce Webinar Hands off the leads it made Run Account-Based Play Reports engagement Each target is read as an early signal for the goal

Each process reports against its own target, and the mesh manager moves the quarter's budget toward the processes that do the most for pipeline. Within a run, the process decides how the work is done. Across runs, the goal decides which work is worth doing.

A process crosses harnesses, so only a mesh can run all of it.

The roles in a real process rarely work for the same vendor. In paid search, the media buyer may be the agency's agent, analytics may be the ad platform's own agent, the landing pages may be built by an agent in the content platform, and legal is a person. Each of those runs in its own harness.

A process built inside one agent framework works when every role lives in that framework. Most companies' processes don't, because the agents they already pay for live in many harnesses. A mesh runs the same written process across all of them: each role is filled from the roster wherever its agent runs, and the gates, limits and ledger hold for every step.

A mesh improves its processes as well as its agents.

A mesh gets better in two ways. Its agents get better at their steps, as Parts 1 and 2 describe. Its processes get more efficient too: the mesh finds the waste in how the work is done and writes a better version of the process.

The work history shows where a process loses time or money: a step where work waits, a check that never fails, an approval that always says yes, a step whose output nothing uses. The mesh manager proposes the fix as a new version of the process, such as a step removed, two steps run side by side or a hand-off rerouted. In Part 1, drafts waited two days at the SEO agent. The fix was a new version of the content process, with the SEO step removed and content handed straight to outreach.

A new version does not go live on the strength of a good idea. It is replayed on past runs first, with no outside effects, and the process's owner approves it before its first real run. Every run records which version it followed, so the ledger shows whether the change helped.

The mesh also learns across processes. Goal telemetry shows which processes move pipeline for each dollar spent, so the mesh manager can, for example, give more of the quarter's budget to a nurture sequence that converts and less to retargeting that doesn't. Part 2 explains how the mesh manager is trained to make those calls.

Start from a process the company already runs.

Goals tell a mesh what to aim for, and processes tell it how the company wants the work done. A mesh needs both: goals without processes leave every agent to invent the work, and processes without goals keep running whether or not they help.

Start with one goal that matters and the processes that serve it. Begin with one of the catalog's 170 starter processes or with how the team works today. Fill in the owner, the limits and the targets, and fill each role with an agent the company already has or a person. The mesh runs them inside the company's rules, records every step in the ledger, learns which processes move the goal, and makes each of them more efficient.