#Sygnals
Signals on AI, operations and the systems companies run on. Opinion and technical.
-
When colleagues can @mention your agent: design the intake before the requests arrive
Microsoft's Autopilot lets colleagues @mention a person's agent like a coworker. Here is how to decide what it may take on, commit to and record.
-
When an AI agent hands work to a person, that person's job has changed
Zoho now ships AI agents that hand customers to a representative with full context. The handoff works. The role receiving it still has to be redesigned.
-
Undo is a design decision: classify every agent action before it runs
Commvault now sells recovery for what AI agents change. A backup restores data, not a sent message. Here is how to design undo into each agent action.
-
Clean data is what a redesigned operation produces, not what it needs first
OpenAI and Databricks now bring ChatGPT to governed company data. Most companies will read that as "fix the data first". The order runs the other way.
-
Self-hosted agents move the server, not the boundary: map what crosses it
Anthropic now lets companies run its managed agent service on their own infrastructure. Here is how to map what still leaves, and what you now have to run.
-
The agents shipped. The operating model did not.
Genpact finds 79% of large companies run AI agents in production, but only 9% have changed their operating model. That gap is where the value goes missing.
-
When your business shows up inside ChatGPT: how to wire the operation behind it
OpenAI now lets companies manage how they show up in ChatGPT. Here is how to connect that new front door to the records, owners and stages your team already uses.
-
A pre-built agent runs someone else's operation
IBM and Microsoft now offer 6,000+ pre-built agents and assets for industry processes. A catalog is a start. What makes you different is still yours to design.
-
A retry is not free: how to stop agents from re-doing work during an overload
New research shows retries keep an overloaded model engine from recovering. Here is how to set retry budgets and idempotency for agents in your operation.
-
The answer your agent needs lives in one person's head
Most companies have one person who knows where the real answer is. An agent cannot ask them. Writing that knowledge down is the first job of any operation redesign.
-
A kill switch needs a safe state: how to design the stop before the agent runs
CrowdStrike just shipped an AI Kill Switch for agents. Stopping an agent is the easy half. Here is how to define what a stopped operation looks like.
-
Speeding up a step does not speed up the work
Most AI pilots make one task faster and leave the queue untouched. The time a case spends waiting between steps is where an operation is won or lost.
-
Cost per resolved case: the one unit metric that tells you if an agent pays off
Token bills and hours saved do not show whether an agent works. Cost per resolved case does. Here is how to define it, measure it and act on it.
-
Ten agents in production, none crossing a department line
IDC found half of companies run 10+ agents but only 3% scale them across departments. Counting agents measures activity, not an operation.
-
Define the term before the agent reads it: a glossary with owners
Snowflake now serves agents only approved business definitions. Here is how to write the glossary your own agents need, with one owner per term.
-
Every tool an agent can call should start switched off
Uber's agent gateway lists every tool as disabled until the owning team approves it. That is operations design, and most companies skip it.
-
Change control for AI agents: treat every prompt, tool and model swap as a release
An agent changes when its prompt, tools or model change. Here is how to version, test and roll back those changes so the operation around it keeps working.
-
A revoked agent permission is a design bug, not a security win
A new survey finds 78% of companies have pulled permissions from an agent already in production. Revoking after the fact means the job was never written down.
-
When another company's agent calls yours: five rules for the handoff
SAP now lets third-party agents invoke its agents through a governed gateway. Here is how to design what your agent accepts, returns and refuses.
-
The exception path is the operation
Every agent demo shows the easy case. Production is the exceptions, and the company that designs where they go owns the real operation.
-
Agent-written code needs its own room: limits to set before an agent runs code
Akamai just launched a sandbox for code that agents write. Here is how to decide what an agent may run, reach and spend before it executes anything.
-
A passing agent eval is not a working operation
Salesforce now ships evals and cost limits for agents in production. They test the agent. The thing that has to work is the operation around it.
-
The ten-step agent: why short, controllable runs beat autonomy in production
A study of agents in production found 68% run at most 10 steps before a person steps in. Here is how to design operations around short, bounded agent runs.
-
Routing models saves tokens. Routing work saves the operation.
NVIDIA's open-source router sends each agent step to the cheapest capable model. Useful, but the bigger savings sit one layer up, in who owns each step of the work.
-
The distrust tax: how to cut the hours your team spends babysitting agents
G2 finds 31% of buyers lose six or more hours a week to overseeing agents. Here is how to design review so oversight shrinks as an agent earns trust.
-
One agent for all your work is a front door, not an operation
Google now offers a single, universal agent for work. A universal agent is a good front door. The operation behind it still has to be designed, step by step.
-
When an agent can work for 24 hours: checkpoints, budgets and handoffs
Notion says its agents can now work on a single task for up to 24 hours. Here is how to design the operation around a long-running agent so it stays controllable.
-
Every AI agent now gets an ID. None of them gets a job.
Microsoft now gives every new Copilot Studio agent its own identity and a human sponsor. An identity says who an agent is, not what work it owns.
-
Ship, scale or kill: the decision rules every agent needs before launch
A public registry now tracks which enterprise AI agents shipped and which were switched off. Here is how to write the rules that decide yours.
-
Write the procedure before you deploy the agent
Enterprise agents keep stalling before production. Build one that lasts by writing the approved procedure first: steps, limits, escalation and a journal.
-
A control tower is not a map of your operation
Vendors now sell governance for AI agents. Governance tells you what agents did. Only a redesigned operation tells them where to work.
-
An agent on a broken system is a faster broken system
Why adding AI agents to the way a company already works rarely pays off, and why redesigning the operation first is the move that does.
-
Anatomy of an operating flow: WhatsApp, CRM and data in one loop
The five parts of an AI operating flow that turns customer conversations into resolved work, and the metrics that tell you whether it is working.