Neural · Professional services
The one we can show you running.
A firm that bills time and delivers projects has a specific problem: the people who know the most are the bottleneck on everything, and the signal that a project is drifting arrives after the month closes. CKS is that kind of firm, which is why this is the sector where we can show you the running system rather than describe it.
Read this first. Neural is the agent architecture we run inside our own company. We run our own delivery on it every day, which makes this the one sector page where most workflows are live rather than designed. We have not run it inside another firm. Every capability below is labelled for exactly where it stands, and the labels separate what runs at CKS from what is in client beta from what is only designed.
Contents
- 01Where it goes wrongThree places, in the order they cost you
- 02One morning, our own firmThe whole loop, start to finish
- 03The workforceFive workflows and exactly where each one stands
- 04What we've actually doneStated precisely, including what we haven't
- 05DeploymentThe questions your IT group will ask first
- 06How it startsOne workflow, sixty days, read-only
[01]Where it goes wrong
Three places it costs you.
The partner is the bottleneck
Everything routes through the two or three people who hold the context. Their inbox is the queue for the whole firm, and the work waits behind whatever they're reading.
Scope creep is unbilled by default
The extra request arrives in an email, gets done because it's easier than the conversation, and never reaches an invoice. Nobody decided to absorb it. It just happened, repeatedly.
You find out at month close
Utilization, realization and the project that quietly went underwater all become visible on the same day, which is roughly three weeks after anything could have been done about it.
[02]One morning, our own firm
This is the loop that produced this page.
This is our own operation, described as it runs.
- 06:00
The brief is written before anyone is awake: what changed overnight, what slipped, what needs a decision today, ranked.
- 08:15
A reply in the thread routes the work. The agent that owns that domain picks it up with its full context intact.
- Through the day
Specifications get drafted, code gets written and independently reviewed, the delivery board stays stocked and ranked.
- Every gate
A senior engineer reviews and merges. Nothing consequential ships without a person, and the platform enforces that rather than trusting it.
- ·
This page, the site around it, and the systems described on it were produced through that loop.
This is the only sector page where the workflows below are mostly live rather than designed.
[03]The workforce
Five workflows, and where each one actually stands.
Every one has the same shape: a trigger, agent work, a human checkpoint, an artifact. The status under each name is the honest part, and it names where a workflow has actually run rather than how finished it feels.
Morning brief
Trigger
Every morning, before the day starts.
Agent actions
Reads calendar, inbox, project state and open commitments, then ranks what actually matters today.
Human checkpoint
A person reads it, and corrections are written back into its instructions rather than into a model.
Output
One prioritized brief.
Inbox triage and drafts
Trigger
Mail arrives.
Agent actions
Classifies, de-duplicates and routes, then drafts replies for the ones worth answering.
Human checkpoint
No email is sent without a person's approval.
Output
A triaged inbox and ready drafts.
Delivery board upkeep
Trigger
Daily, as the board drains.
Agent actions
Validates queued work against a quality gate, ranks it, and promotes what is genuinely ready.
Human checkpoint
Humans set the gate. Anything ambiguous is flagged rather than forced.
Output
A stocked, ranked board.
Ops watch
Trigger
On a recurring schedule.
Agent actions
Audits infrastructure, dependencies and exposure, and verifies that backups actually restore.
Human checkpoint
Findings are reported, never silently fixed.
Output
A written audit with recommendations.
Scope drift watch
Trigger
As requests arrive against an engagement.
Agent actions
Compares what is being asked for against what was scoped, and names the drift while it is still a conversation.
Human checkpoint
The engagement lead decides whether it's absorbed or billed.
Output
A drift note with the original scope line quoted.
[04]Standing
What we've actually done.
Stated precisely, because in this category precision is the differentiator.
This one isn't a projection
Four of the five workflows above run CKS itself, every working day. The brief that starts our morning, the triage on our inbox, the board that feeds our engineers and the audit that watches our infrastructure.
It produced work you can inspect
A maritime safety app live on Google Play, a construction platform in client beta, and this website. All built through the loop described above, with senior engineers owning every merge.
We'll show you the console
Not a demo environment. The system we actually run, with its real board, its real audit log and its real approval gates, on a call.
What's different at your firm
These workflows run at a five-person firm. We have not validated them at a larger one, where permissions, segregation of duties, multiple practices and far higher exception volume all change the problem. A pilot tests workflow fit and integration, not just plumbing.
What we haven't done: run this inside another firm's systems. Dogfooding proves the workflows operate; it does not prove they transfer to a firm with more people, more practices and stricter segregation of duties..
[05]Deployment
The questions IT will ask.
Answered here rather than in the third meeting.
- Where it runs
- Your infrastructure or a private tenant. Not a shared multi-tenant product.
- What it reads
- The systems you already run. Read access first, always.
- Who approves
- Actions you designate as consequential stop at a named person before they happen. You set that boundary; the platform enforces it rather than relying on discipline.
- What's recorded
- Each agent action is logged with its inputs, the sources it cited, what it did, and who approved it. The things that hold up in a review.
- Who owns it
- Your data and configuration are yours, and we'll put the exit terms in the contract rather than asking you to trust a web page.
[06]How it starts
One workflow. Sixty days.
Not a platform rollout. We pick the single workflow where it is costing you most, stand it up against your real systems, and you judge it on whether it caught things your people did not have time to.
Week 1
The workflow and the success metric are agreed in the first session, not discovered over a month. Read-only access to the systems it needs; your security review runs in parallel.
Weeks 2–7
It runs alongside your people on live work. Every output reviewed by your team, nothing acted on automatically, false positives tuned weekly.
Week 8
Measured against the metric we set in week 1. Keep it, change it, or walk away. Because it stayed read-only, walking away means revoking credentials and removing an integration rather than migrating off a platform.
- What it costs
- Fixed fee, quoted after the working session, so it is priced against your actual systems rather than a guess. No per-seat licence during the pilot.
- How it's judged
- One number, agreed before we start. Usually exceptions caught that your team didn't have hours to catch, measured against a baseline period.
- What we need
- Read-only access to two or three systems, one person who reviews the output, and an executive who reads the weekly.
- What you keep
- The configuration and everything it produced, whether or not you continue. Data handling and retention are in the agreement before access is granted.
A working session is ninety minutes with your operators, not a pitch. We map where the cost actually lands and tell you which parts of it are worth pointing an agent at, including the parts that are not.
Other sectors