Jungle Expedition, month 4: what we shipped, earned, and corrected
The first edition of our monthly operating report. Ten sites in production, one new client, about 2,500 euros in revenue, a 22.6 percent outbound reply rate, and the channel we ignored for four months.
Every fourth Sunday we publish how this company is actually going. What we shipped, what it earned, what we corrected. This is the first edition and it covers July 2026.
AI Jungle LLC opened on April 1st. We install and operate AI agent workforces inside small consulting firms, and we run on the workforce we sell: our own sites, content and client operations are run by the same kind of agents we install for clients.
Where does the idea come from?
Not from the technology. It comes from ten years of moving technology between Europe and Asia, and from watching the same trap close on firm after firm.
Small cross border consulting firms live off the partners' brains and die of the operational work: the CRM, the monitoring, the follow ups, the content. They are too small to hire and too specialised to outsource. Beautiful margins, clients who depend on them, and no way to grow without becoming a Big 4 and inheriting its staffing and quality problems.
A properly installed agent system is the first employee such a firm can actually afford. We built it for ourselves first. Then a contact said: install me the same one.
What are the five rules the business runs on?
1. Business first, not AI first. Most companies selling agents are AI companies looking for a business that needs them. We start from what caps the firm, which is almost always the owner's hours, and only then decide whether an agent is the answer. Our buyers are not technophiles. They run consulting firms, and they spend the day on the phone rather than in front of a screen. A director does not want to operate a dashboard. He wants an employee.
2. The best of three continents. European standards on process and privacy, Asian discipline on speed and cost of execution, North American clarity on how an offer is packaged. That is not a line about being global. Our entire AI stack costs about 575 euros a month, everything included, and one rule keeps it there: subscriptions only, never a metered API key left running inside an autonomous loop. A billed key on an agent that works all night is how a bootstrap dies in a weekend.
3. Their results are our results. We do not count deliverables, we count what changed for the client.
4. Price on results, so both sides sit on the same side. A meaningful share of what we earn is indexed on client performance rather than on a flat fee. It removes most of the friction in a price negotiation, and it is also a bet: some months we earn less.
5. Sell the whole arc, not the delivery. This is the one we underestimated, and it is where most agent projects quietly die. Five stages, described below.
What are the five stages of an engagement?
| Stage | What happens | Why it exists |
|---|---|---|
| Diagnostic | We name the bottleneck and what it costs to leave it there | A technology assessment tells you nothing about the business |
| Test | The hypothesis goes into a demo on the client's own data, fast | So nobody discovers they were wrong three months and one invoice later |
| Build | The agent enters the real workflow, with approval built in | Nothing ships without your yes |
| Training | The team learns to run it | Most vendors skip this, and it decides whether the system survives month one |
| Improvement | Two loops running at once | A system that stays where we left it decays |
The inner loop is the agent improving: every correction the owner makes becomes a rule it follows from then on. The outer loop is the team improving: working with an agent forces a team to describe its own standards precisely, and those standards feed back into the system. The firm ends up owning its operating knowledge, written down and executable. That is what Transfer of Experience means in practice.
One loop alone decays. An agent that improves inside a team that never learns becomes a black box nobody can question. A team that learns without a system to capture it just knows more and stays as slow.
And the loop that matters most is not automated at all. We hold a weekly call of at least an hour with every client, whatever they signed.
What did a client actually get?
Asia Connect is a firm specialised in interim management and in solutions that are not business as usual in Asia. They could not add a business development person, and they did not want to.
Over three months: 755 executives contacted, 1,992 messages sent, a 24% reply rate, 139 qualified meetings booked, and 90 of them held.
Zero messages sent without human approval. Not a single one.
That is their result, and it is the only version of ours that means anything.
What did we ship in July 2026?
Ten site rebuilds between July 12 and July 21. Production, not mockups. Seven of the eight chats on those sites are real agents with memory and their own profile, each tested against prompt injection. Ask one for its system prompt and it will refuse.
- A first verified third party review, from our first client, with a named quote cleared in writing.
- Pitstop taking real money end to end, from signup through checkout to fulfilment.
- The Leverage Score live: ten questions, a deterministic score out of 100, no call required.
- Our agent machine format open sourced, so anyone can see how we package agent work.
- A full exit from our previous hosting provider. Ten sites out of ten on infrastructure we control, with server memory down from 9.2 to 8.0 GB.
- Twelve countries of pet relocation rules researched and published on one of our sites, every figure backed by a government source.
Infrastructure and content are the same job here. A boutique operating ten brands is only possible if the cost of running the eleventh is close to zero. The ecosystem page lists what we run and which clients agreed to be named.
Where did month 4 land?
- One new client signed, starting in September
- About 2,500 euros in revenue, variable by design because it is indexed on client performance
- A 22.6% reply rate across 2,088 outbound messages
- Ten sites in production, eight live agent surfaces
Small numbers, and results-aligned pay is exactly why month 4 looks like this. Same lines next month, and you will be able to compare.
Which channel have we been ignoring?
For four months we built and barely distributed. The last four weeks say so plainly: 7,766 impressions across 124 posts on X producing 8 link clicks and no new followers, 582 impressions on our company page producing 3 visitors, and 1.3k views on a single comment left in a discussion forum where our buyers already are.
Our best piece of distribution was a comment on somebody else's post. Not a launch, not a film. A useful answer in the right place.
So distribution becomes a real activity in August, and we will publish the same three rows next month whether they improve or not.
What gets graded next month?
Three commitments for August, graded A to F in the next edition:
- A first inbound qualified conversation from a site we rebuilt, rather than from a warm introduction.
- Every automated job reporting the number it produced, never only a success code. For any job whose purpose is to produce something, a green status tells you the machine ran and nothing about whether the work happened. Google's site reliability practice has argued for years that monitoring should be built on symptoms users experience rather than on process completion.
- The new client's kickoff prepared and scheduled, with no reopened price.
Publishing the forecast before the month is what makes the grade mean anything.
How does the referral program work?
Send us a firm and if it becomes a client you receive 10% of what we actually collect on the install and the monthly retainer, for their first 12 months, paid after we have been paid. It does not apply to the performance share, because that part is the alignment with the client and it is not ours to hand out. The firm you send gets 10% off as well.
Two conditions, neither negotiable. Register the firm before the introduction, and we confirm within two business days whether they are already in our pipeline. And disclose that you may be paid, clearly, every time you recommend us publicly, as the FTC Endorsement Guides require. We disclose it to the client before signature too. Full terms are on the referral page.
Where should you start?
If your firm runs on your hours, the Leverage Score takes about three minutes. You get a number out of 100 and the place where your ceiling actually sits. No call, no calendar.
Next edition: the fourth Sunday of August, with the same lines filled in.