The thesis#

The economics of consulting are shifting. The traditional structure of a senior partner, four associates, eight weeks of work, and a sixty-page deliverable was a function of how expensive it was to produce structured analysis at scale. That cost has dropped sharply with the current generation of language models and agent frameworks.

The competitive position that follows is straightforward: the consultancies that win the next decade will be the ones that ship the first repeatable, AI-powered version of each major deliverable. Productization, not headcount, becomes the differentiator.

This is the bet Ai Solucija is testing. The shipyard sector was selected as the first vertical for four reasons:

  1. The buyer universe is bounded and findable. Every shipyard in a given region can be named and scored. There is no diffuse long-tail problem.
  2. The deliverables are well-defined. Tender reports, capability dossiers, bid analyses, and board briefings have stable shapes across customers.
  3. The current work is overwhelmingly manual. Tender flow is scattered across dozens of public procurement portals, often in multiple languages, and the prevailing tracking method at most yards is "we will hear about it from a broker."

What is being built#

Three product lines are under active development, listed roughly in order of maturity.

1. Tender intelligence platform. Twenty-nine country-level scrapers feed a weekly digest pipeline. The most recent test run processed 922 raw tenders, validated 285 against shipyard-relevance criteria, and produced a single recommendation report. The technical challenge is not the acquisition layer but the filtering, multilingual classification, and quality assurance: specifically, catching the failure modes that would otherwise reach a procurement team. This product is the closest to a recurring-revenue model.

2. Capability matching and proposal support. Given a public tender and a structured shipyard capability profile, the system produces a go/no-go recommendation in approximately four hours, compared to the two- to three-day cycle typical of manual analysis. Where the recommendation is positive, it generates a three-document proposal package: formal application, filing instructions, and a justification narrative. This is a per-bid service offering.

3. Thought leadership briefings. Long-form OSINT-grade analyses of naval programs: the kind of in-depth documents that a yard's board reads in advance of significant capital decisions. The current production process is a manual research effort; the rebuild is an AI-assisted pipeline with a three-stage quality assurance loop (factual verification, language quality, brand voice).

Each of these has working artifacts. None of them has a paying customer.

Why publish before selling#

The conventional consulting sequence is: secure a client, deliver the work, publish a case study after the fact, and accumulate brand equity from delivered outcomes. The reasoning behind the opposite sequence (publishing first) comes down to three observations.

First, in this market the artifact is the proof. A complete long-form naval program analysis with consistent typography and verified citations demonstrates capability more credibly than a logo wall. Most readers in this niche would not recognize a logo wall in the first place. The output is the credential.

Second, productization itself is the differentiator. A consultancy that cannot articulate its delivery process (that produces wet clay every engagement) has not productized anything. Publishing each component publicly forces the underlying process to be defensible, repeatable, and explicable. The discipline of the journal is the discipline of the product.

Third, build-in-public is the most efficient distribution channel available at this stage. The audience is small but specific: shipyard operators evaluating new capabilities, fellow consultants navigating the same transition, and operators in adjacent industries who will recognize the pattern. Reaching that audience through paid channels would be inefficient. Showing the work directly is not.

Scope of the journal#

This is a weekly account of work in progress. Each entry will cover what was built, what was abandoned, and what was learned, with concrete technical detail where it is informative.

Three commitments shape what will and will not appear:

  • No thought leadership without substance. Posts that do not document a real piece of work that week will not be published. Trend pieces, predictions, and abstract commentary are out of scope.
  • No client identification. Methodology, code, and numbers drawn from internal datasets are publishable. Anything that could be reverse-engineered to identify a specific shipyard, individual, or engagement is not. Anonymization is applied before each publication.
  • No tutorial framing. Claude Code and related tooling appear throughout the work. The objective is not to teach those tools, but to demonstrate what becomes possible when AI is used to package consulting deliverables rather than to build applications.

What follows#

The work in front of me sits across several distinct strands: the scraping architecture, multilingual quality assurance, the consulting menu structure, and an extensive recent cleanup of the codebase. Subsequent entries will draw from that work as it develops. The cadence is one substantive post per week, published Friday or over the weekend.

For shipyard operators, fellow consultants, or operators in adjacent industries who would like to discuss any of the above: direct contact is welcome via LinkedIn or email.

Tit Dolinšek, Ai Solucija