Find, Open, Create, Help: What Arrives with PMWEB 2026

Ask a project director why the data in their PMIS is thinner than it should be and you will rarely be told that people do not understand the system. You will be told that the people who have the information do not enter it, and the people who enter it do not have the information.

That gap is usually described as a training problem. It is not. It is a friction problem, and it is worth understanding precisely, because the first AI capability arriving in PMWEB 2026 addresses exactly that and nothing more ambitious.

Four behaviours, available wherever the user happens to be working: find, open, create and help.

Adoption is a data completeness problem before it is a training problem. Every workaround a user adopts removes data from the system of record.

The cost of knowing where things are

A configured PMIS is a large application. Modules, sub-modules, a record tree, filters, saved views. An experienced user navigates it without thinking, which is precisely why the cost is invisible to the people who evaluate the system.

The occasional user is the one who matters here. The site engineer who raises four records a month. The commercial manager who needs one figure before a meeting. The consultant who was given access for a specific purpose and uses it fortnightly. For each of them the navigation is not automatic, and the cost of working out where something lives is high enough, often enough, that the rational response is to do it another way. They send an email instead. They ask someone. They keep it in a spreadsheet and enter it later, or not at all.

PMIS does not reflect the programme as fully as it should.

Find

The user describes what they need in plain language and the assistant searches records across PMWEB, returning linked results.

The important qualifier is that it searches what the user is authorised to see. This is not a permission layer bolted on afterwards; the assistant uses the same access control that governs every other part of the platform, which means it cannot be configured into an inconsistent state and cannot be accidentally bypassed. For an organisation running segregated commercial teams across programmes, that inheritance matters more than the search itself.

It is also the second question on PMWEB's own AI checklist for buyers: does it respect the permissions already in place? The Intelligence Control Framework answers it in the terms that matter, which is that AI works under the permission model the owner already administers rather than a second one.

Open

The assistant takes the user directly to the best-matching record, removing the traversal through the module tree entirely.

This sounds like the smallest of the four and it may be the one that changes behaviour most, because it removes the specific friction that stops occasional users. The user no longer has to know where a record type lives in order to reach an instance of it.

Create

An authorised record or adaptive form can be started from anywhere in the platform, without navigating to the from it. In the demonstration of a commitment change order, eight of twenty-one fields were pre-filled from context before the user typed anything.

Eight of twenty-one is worth sitting with, because it is a more honest number than a percentage would suggest. It is not most of the record. It is the portion that could be derived from where the user already was, and it is the portion that is tedious to retype and easy to get wrong. The remaining thirteen fields are the ones that require judgement, which is the correct division of labour.

The draft opens in a separate tab, so nothing in progress is disturbed, and it is not saved until the user reviews the fields and saves it. As covered in an earlier article in this series, that gate is a design commitment rather than a limitation of an early release.

Help

Contextual guidance, suggested prompts, follow-up questions and access to previous assistant sessions, delivered inside the working context.

Product help has historically lived somewhere else — a manual, a portal, a training session six months ago. Help that arrives where the user is stuck is a different proposition, and for a platform whose adoption depends on occasional users it may be worth more than the three capabilities above it.

Why this tier matters more than it appears to

None of this is insight. There is no analysis here, no early warning, no recommendation. It is a set of capabilities aimed squarely at the friction that stops people entering data and finding it.

That is exactly why it is the right thing to ship first.

Every capability further up depends on the data being there. Everything else in AI for project systems is downstream of that.

An organisation that deploys sophisticated analysis on a thinly populated PMIS gets sophisticated analysis of an incomplete picture, delivered with more confidence than the underlying data deserves. Removing the friction that keeps the record incomplete is not a modest first step ahead of the interesting work. It is the precondition for the interesting work.

One commercial point worth noting

Several platforms are introducing usage allowances for AI capability — token budgets, query limits, per-seat entitlements. PMWEB's stated position is that AI usage carries no limit.

For a capability of this kind that is not a minor commercial detail. Capability aimed at occasional users only works if occasional users are allowed to use it freely. A team told to ration something does not adopt it; it reverts to the email and the spreadsheet, and the friction problem returns intact.

What to do about it

Two practical points for organisations already running PMWEB. This capability requires PMWEB 2026, so establish your current version and your upgrade path now rather than at release. And AI capability can be enabled individually, which means a phased rollout is available to you — worth planning deliberately rather than switching everything on at once.

For organisations not on PMWEB, the thing to take from this is the sequencing. When a vendor shows you AI analysis before they have shown you AI that makes the system easier to populate, ask what the analysis is going to be reading. And ask to see it run on your live records rather than a prepared sample. PMWEB's own checklist makes that the last of six questions to put to any vendor, including itself.

Everything else in AI for project systems is downstream of the data being there. Removing the friction that keeps the record incomplete is not a modest first step. It is the precondition for the interesting work.

Next week: turning a signed contract into a structured record, and why showing the source matters more than the extraction.