Hail, Orchestrator

How to build a digital company today, AI-first.

I loved the world of legacy business.

Getting dressed up… The anticipation of a meeting… Sharing jokes with your team as you work towards a shared goal. Yet, there were times of silence in a gloomy office trying to focus.

But Zoom, and office towers and meeting rooms will feel a curse after reading this.

In hindsight, this 20th century phenomenon of clustered desks around basic human amenities like water and snacks… is starting to feel like a momentary aberration… like COVID lock downs, smoking indoors, and 2000s ultra-skinny jeans.

Of course, five-day, office-first knowledge work was disrupted by the forced adoption of work from home culture in 2020. But with the corporate world busily trying to force the toothpaste back in the tube…. A larger disruption was coming with the advent sufficiently smart AI for some of us to become work agent-first. And it’s spreading the toothpaste everywhere.

The notion of humans making decisions via Zoom calls and conversation, using ‘complicated airflow’, is starting to seem archaic to anybody taking an agent-first approach to building a business.

When you get used to OpenAI or Anthropic agents giving you PhD level consultation at warp speed, the idea of asking a human professional expert is unpalatable, and a major competitive disadvantage. The cost, the time, the politics, the entrenched interests…

Humans as liability

You’ve heard the criticisms before… they eat, sh*t and need sleep. I think it’s worse than that. Humans are painfully slow and protective… political… and make mistakes.

It is a liability today to wait for accounts, marketing, branding, tech (or any discipline really) for their opinion or approval, waiting for them to ‘get back to you’ when you have the answer in an agent is bizarre.

There are times when these professionals might be better than an AI. By how much? 5% ? 10% ? Is it worth it to employ a human for that? At what cost?

Human Resources. Even the term itself always eluded a wry smile. The thankless task of encouraging maximum productive output and the compliance of humans… A notoriously chaotic breed.

Whilst legacy businesses have multiple personalities competing to show their little bit of value transmitting information through the noisy and inefficient albeit beautiful method of human speech or the forced formality of corporate email, there are machines making the same decisions between themselves in seconds. And doing a better job.

Anybody that has run a company or team of any size knows that the people are key…. finding and nurturing the good ones, motivating them to work, hard and make smart decisions aligned with the success of the company was a noble (and sometimes rewarding) pursuit.

But that skill is increasingly unnecessary.

The evolution of the Computer

The word computer was originally a human job title for those performing calculations, often using tables, mechanical calculators, or pencil and paper.

Machines eventually took over both the profession and the name.

Humans doing this work was a necessary insanity, and we all know that spreadsheets were the method for the last thirty years.

Well, that line of digitisation is moving again.

With the advent of generative AI and frontier models, we have now crossed over into a world where the scope of the work done by humans should be so dramatically reduced as to render them simply visionaries and arbiters of taste and authentication.

Nobody now would think to do their finances by hand when a spreadsheet is available.

Well today, people working agent-first would not think to assign work to a human before an agent.

If you spend enough time working this way, you notice that your role becomes instead orchestrating a relentlessly efficient array of vastly intelligent agents, with the main pauses being the maching verifying decisions and asking for authentication.

Enter the Orchestrator

An Orchestrator is gifted generalist that can orchestrate multiple agents across different disciplines. Their competencies are defining vision, refined taste and being organized enough to get the most out of that agent team.

And (for now), they are human beings.

A tech founder today is an Orchestrator.

They use a swarm of agents for all the traditional functions you would find in a software company… Engineering, Brand & Marketing, Sales, Accounting, Compliance.

‘Human-bound’ exceptions

Some types of work remain human-bound.

• Legally human-bound — protected or accountable roles: lawyers, accountants, directors, licensed professionals, signatories.
• Physically human-bound — manual labour. We don’t yet have widespread functional cheap robots… for now.
• Socially human-bound — roles that depend on trust, persuasion, rapport, status, empathy, or interpersonal legitimacy. Sales is the clearest example.
• Economically human-bound — work AI could theoretically automate, but where a human remains cheaper or easier for now; normally overlaps with Physically human-bound.

You should be designing companies AI-first before identifying these ‘human bound’ exceptions. And it goes without saying that these humans should be using AI to augment their work where possible.

So who do you hire in an agent-first company?

Beyond human-bound roles… Who does a founder hire today to speed up the growth of their company?

More orchestrators of course.

In the past, where your first hire may have been a gifted product owner, operations manager or industry expert, you now seek out someone with orchestrator qualities. Someone that is a talented generalist that can take initiative and keep their eyes on the goal.

These people will not be cheap. But they will be incredibly productive.

Practically, how does this work?

I’m using OpenAI as my orchestration stack. (ChatGPT product names are confusing, but I explain them here.)

Where I would hire a person or a team, I create an agent.

Let’s say I’m building a company “Pineapple”, I create the Codex folder “Pineapple” with something like…

• Leadership
• Growth / Marketing & Brand
• Engineering
• Design / User Experience
• Sales & Partnerships
• Finance / Accounting

An example prompt for a leadership agent would be:

You are the Leadership function for Pineapple, working directly with me as the founder.

Your role is to help me run the company: shape strategy, set and refine goals with me, establish priorities, identify the highest-leverage opportunities and risks, and coordinate work across product, engineering, compliance, growth, marketing, operations, and commercial strategy.

Do not assume that you independently determine the company’s goals. I will often set, change, or help define the objectives. Your job is to challenge my thinking where useful, surface trade-offs, propose better approaches, and turn our agreed direction into coherent priorities for the rest of the organization.

Maintain the whole-company view. Specialist agents should own deep execution within their domains; Leadership should synthesize their work, identify conflicts or gaps, challenge assumptions, and help decide what the company should do next.

Act like a strong founder’s leadership team: commercially ambitious, technically literate, customer-focused, financially pragmatic, candid about uncertainty, and willing to bring me ideas rather than simply waiting for instructions.

Eventually I realized my own role was closer to a co-CEO or maybe even a board member.

That Codex folder that belongs to the product repo. The repo becomes the entire source of truth for the business with everything from the governing principles to nuances of tone of voice, all documented in the product’s ‘docs’ folder. That way every agent is aligned with the company. If you’ve worked with humans… You’ll know how hard this is.

Closing thoughts

I plan to write more about this new idea and I am eager to get feedback.

Velocity

I think it’s obvious to anybody building software that what would have taken two years could potentially be done in a weekend.

I used to get so caught up with small decisions.

Those are instantly unblocked. e.g exporting videos for my own socials

Existing business

If you have any employees at all beyond the ‘human bound’ exceptions above, you will find this transition difficult. And that will turn out to be a competitive disadvantage.

“AI makes mistakes”

Anybody relying on this today is either insecure about their profession or just is simply not using the tools. AI does make mistakes… but Dramatically less than any human I’ve worked with, and with much broader skills. Lawyers… pay attention. Only your title protects you.

Working alone

I’m writing and posting a lot more simply because… The consequence of all this is you spend a lot more time alone if you are a small entrepreneur.

Jobs market

I really don’t know, so don’t ask me. Plenty of people are willing to share their opinions on this topic. I do think the Orchestrator role will become a thing, In the same way Product Owner was popular in the 2010s.

What next for legacy business?

Medium and large corporations are going to be around for a long time. Despite all the excitement about AI… My favorite airline still asks me for the same information every time I book. More importantly, organizations are human political constructs with entrenched interests that will preserve the status quo for a while yet. In many industries this will definitely mean opportunities for disruption from AI-first companies.

What’s your experience?

Are you building a company like this now? Or multiple companies? What did I miss?

Send me ideas…

Do you have a software company idea? Or just a fun project.

I have time and tokens and I’m feeling inspired.

Contact me

Discuss in a comment or email me directly gm@web2boomer.com

Written by a human hands ;)