Home / Software & AI / Document Intelligence / Agentic Automation

AI agents for the uncertain steps — rules for the rest

Either someone has sold you agents and you cannot tell what makes them different from automation, or your pilot works in the demo and fails internal audit. Both questions have the same answer, and it is about which steps can survive a system choosing its own sequence.

from $30,000Solution Blueprint: step-by-step agent-or-deterministic decision, in 5–6 weeks
from $150,000typical build — hybrid pipeline with intent-level logging, 4–6 months
Fixed pricequoted against the Blueprint scope — no time and materials
On-prem or cloudpinned open-weight models inside your perimeter, GPUs in the same contract

This is for you if

In your words rather than ours.

You are being sold agents and cannot tell what is newEvery vendor deck says agentic. Nobody has explained what it changes about your process, your costs or your audit, and the demos all look the same.
The pilot works and internal audit will not pass itIt performs on the sample. Then someone asks whether the same file always produces the same result, and the honest answer is no.
Someone disputes an outcome and asks how it was reachedA losing bidder, a supplier on a rejected invoice, a contractor on a flagged defect — or, in the strictest case, a regulator. “The model chose to check this next” is not going to be a sufficient answer to any of them.
Your exceptions genuinely resist rulesYou have tried. Every new rule catches one case and misses the next, because the resolution depends on what turns out to be wrong.

Where we put an agent, and where we will not

This is the whole page. An agent decides what to do next; most processes require the same input to give the same output — not only under a regulator, but in tendering, inspection and invoice matching, wherever someone can disagree with the outcome and ask how it was reached. That conflict is structural and no prompt resolves it, so autonomy gets spent where it buys something and withheld where it costs you the answer.

Agent
  • Exception resolution — the path depends on what turns out to be wrong, and nobody can enumerate the cases in advance.
  • Context gathering across systems — the next lookup depends on the previous answer.
  • Drafting for review — a named person checks it before it has any effect.
  • Routing with a stated reason — the reason is the deliverable and the routing is reversible.
Deterministic
  • Anything that must be reproducible — same file, same answer, permanently.
  • Anything that gets contested — by a bidder, supplier, contractor or regulator.
  • Final actions where the outcome is disputable — payment, approval, decline, release, award.
  • Anywhere the error cost is asymmetric, because a variable path makes the tail unbounded.

One test resolves most cases: does a person check the output, or is the cost of an imperfect one low? If neither, the step is deterministic. The full argument, with what to log →

When we tell you not to do this

Given eight of our competitors have built a content practice around agents, it is worth being explicit about what we will not sell.

An agent as a substitute for a written processIf the process is undefined, an agent does not define it — it hides that it was never defined, and the first audit finds out. Write the process; then decide which steps deserve an agent.
Autonomous final action on a disputable outcomePayment, approval, decline, release, award. We will not build a system that takes these without review — regulated or not, because the person on the receiving end can always ask why.
Agents where rules already workIf the path is writable, writing it is cheaper to run, fails loudly instead of silently and explains itself for free. Replacing working rules with an agent is a downgrade wearing a newer name.
A pilot with no ground-truth setWithout an agreed labelled sample, “good enough” is an argument between opinions and the project stalls at the approval gate. That set is the first deliverable, not a later one.

The architecture we build

A deterministic skeleton with agents confined to individual steps. Three constraints make it hold, and each is a limitation rather than a feature.

Deterministic skeleton — every transition is versioned code intake → classify → completeness → validate → [ AGENT ] → route → decide → record ↑ exception resolution only fixed inputs · schema-constrained output cannot skip a downstream step cannot write outside its boundary intent logged, not only the answer

It cannot change the process

The agent runs inside one box. It cannot decide that validation is unnecessary today, because the sequence is not its to choose — only the work within its own step.

Its output is schema-constrained

A structured result the deterministic layer validates before acting on, not free text handed downstream. Failing validation is an exception, not a silent pass-through.

Its intent is logged

Which steps it chose and on what basis — not just the answer. That is what makes a decision reconstructable eight months later, and it is the part everyone omits until an auditor asks.

We buildWe build secondWe decline
The deterministic skeleton, agents scoped to exception resolution and context gathering, schema validation at every boundary, intent-level loggingDraft generation for human review, routing with stated reasons, cost-per-case monitoring, a ground-truth set and evaluation harnessAn agent in place of a written process. Autonomous final actions on outcomes someone can dispute. Agents replacing rules that already work.

Four regulated processes, one pattern

The reason we are confident about where the line sits is that we have shipped the same shape under four different rulebooks. Client names withheld under NDA. See full case studies →

Immigration tech · four production LLM products

Drafting that a case manager signs

A retrieval-augmented pipeline that reads large, inconsistent document sets and drafts structured memoranda with citations back to source, plus LLM-assisted triage and next-step suggestions wired into the existing case-management system. The model drafts; the case manager decides. That division is the reason it reached production rather than staying a demo. Read the case →

80% of routine drafting automated−45% case processing time+30% throughput
Aviation · MRO

Advisory checking under maintenance rules

Page classification, signature and stamp detection, empty checklist cells and cross-pack number agreement — deterministic checks producing an annotated report, with a specialist signing off. No autonomy anywhere near the sign-off. Read the case →

6 document types4 classes of checkCLI + API
Lending · identity

Verification that hands off rather than decides

Document authenticity, face match and liveness as a separable service that verifies an applicant and passes them to the process that decides — the same preparing-versus-deciding boundary, drawn in a different regulation. Read the case →

−75% fraudulent applications+35% conversion

The fourth rulebook is GMP, where the same completeness-and-signature mechanic applies to pharmaceutical batch records — covered in AI batch record review.

What was hard

On the visa platform the difficulty was not drafting quality, it was deciding what the model was allowed to assert. An early version produced fluent memoranda that stated things the source documents merely implied — plausible, useful-looking, and exactly the failure mode you cannot ship into a process where a person signs underneath. The fix was not a better model but a narrower contract: every claim carries a citation to a specific source passage, and anything without one does not appear in the draft at all. That constraint cost some fluency and is the reason the system survived review.

No surprises

Your data

The stack runs on-premises or air-gapped, with no document leaving your network and GPU hardware quoted in the same contract. Details in security and compliance and private AI infrastructure.

Your reproducibility

Self-hosted open weights with pinned versions, because a re-run only means something if the model that produced the original answer still exists. A hosted endpoint can move underneath you and take your audit trail with it.

Your budget

Fixed price against a scope agreed before the build starts — including the per-step decision on where an agent goes, which is the thing that determines cost per case. No time and materials.

Frequently asked questions

How is an agentic automation project scoped and priced?

Scope first, price second, and both numbers are published. The specification is a product: the AI Solution Blueprint, one system, 5–6 weeks, from $30,000, credited in full against the build if implementation starts within 90 days. Its main output here is the step-by-step decision on where an agent goes and where the process stays deterministic, plus a fixed price and a timeline. Builds of this kind typically start around $150,000 over 4–6 months.

Our agent pilot works in the demo but fails internal audit. Why?

Because the demo never asked the question audit asks. An agent selects its own sequence, so the same input can take a different path and produce a different answer, and internal audit wants reproducibility plus the ability to reconstruct a specific decision months later. No prompt fixes that — lowering temperature narrows the distribution without making the sequence deterministic. The fix is architectural: put the agent inside one bounded step of a deterministic process rather than in charge of the process.

Where will you actually put an agent?

Where the path is genuinely unknown in advance and a person checks the result: exception resolution, gathering context across several systems where the next lookup depends on the previous answer, drafting for review, and routing with a stated reason. Everywhere else stays deterministic, because rules are cheaper to run, fail loudly rather than silently, and explain themselves by construction.

Will you build an agent that takes the final action?

Not where the outcome can be disputed, and we will say so before the proposal. Payment, approval, decline, release and award are the actions you have to be able to explain case by case — to a regulator in the strictest case, and far more often to a supplier, a bidder or a contractor, and an autonomous sequence cannot promise that the same facts would have produced the same action. If you want that anyway, it becomes an explicit scope item with its own risk discussion rather than something folded quietly into a document project.

What do you log so a decision can be reconstructed?

Inputs as received, the model and prompt versions in effect, the intent trace — which steps the agent chose and on what basis, not merely the final answer — and the human review step with the reviewer's identity and what they were shown. The third item is the one usually missing and the one that turns a stored answer into a reconstructable decision.

Can it run on our own infrastructure?

Yes, and for agentic work the argument is stronger than usual. Self-hosted open weights let you pin the model version, which is what makes a re-run meaningful; a hosted endpoint can move underneath you and take your reproducibility with it. The whole stack runs on-premises or air-gapped with the GPU hardware quoted in the same contract.

Related practices

Tell us which step you think needs an agent

An engineer replies, not an account manager. If rules would do the job, you will hear that first.

Want the spec first? AI Solution Blueprint — from $30,000, credited against the build.

We reply within one business day. Prefer email? sales@haink.org