Experienced Engineers

AI Is Automating Parts of Detailing. Here Is What It Cannot Do

Tekla 2026 automates drawing production, connections and validation. An honest look at which detailing tasks are going away, which are growing, and what to learn next.

By Mr Peter D'Souza 22 September 2026 · 7 min read

Every detailer who has watched a software demo in the last year has asked some version of the same question, usually privately: if the software generates the drawings, what exactly am I being paid for in five years?

It deserves a straight answer rather than reassurance. Parts of this job are being automated, and some of them are the parts that junior detailers are hired to do today. Other parts are becoming more valuable, not less. This post separates the two, using what the software actually shipped in 2026 rather than what a vendor video implies, and sets out what to learn if you intend to be on the right side of the line.

What actually shipped

Trimble released the 2026 Tekla software on 11 March 2026. The features worth reading carefully are these (Trimble):

AI cloud fabrication drawings in Tekla Structures, which generate fabrication drawings using drawing libraries built from your past projects.
Simultaneous model and drawing editing, so a change in one appears immediately in the other.
A Layout Manager "out of tolerance" workflow that flags discrepancies between the design model and as-built data.
A project settings management console for administering standardised environments centrally.
In Tekla Structural Designer, integrated portal frame design to Eurocode, and a more accurate centre of rigidity calculation.
In Tekla PowerFab, pre-shipment validation that catches incomplete shipments, and LEED reporting that flags material origin and recycled content automatically.

Read that list again as a detailer rather than as a buyer. Almost everything on it targets repetitive production and cross-checking: drawing creation from templates you have already used, keeping two representations in sync, comparing a model against measured reality, standardising settings across an office, and reconciling what is about to leave the yard.

That is a fair description of a large share of the hours in a typical drawing office.

What is genuinely going away

Be honest about this, because pretending otherwise does not help anyone plan a career.

Repetitive drawing production. Producing the hundredth similar single-part drawing on a repetitive PEB has been shrinking for a decade through templates and automatic drawing creation. Library-driven generation accelerates it.

Manual reconciliation. Comparing a BOM against a drawing set, or a shipping list against a model, is exactly the sort of comparison software does faster and without fatigue.

Standard connection selection on routine geometry. Automatic connection tools already place and dimension familiar connections on familiar members. Their coverage keeps widening.

"Drafting-only" roles. The job that consisted of tracing an engineer's sketch into a model, with someone else deciding everything that mattered, is the most exposed role in the office. If your daily work is entirely instruction-following in the software, that is the risk to take seriously.

What is not going away, and why

1. Deciding what the inputs actually mean

Detailers rarely receive a complete, consistent, final design. They receive a design that is 80% settled, a set of client comments, a revision that contradicts an earlier one, and a verbal instruction from a project manager. Turning that into one coherent model is judgement, and it is the first step of the job. Generation tools work forward from decisions; they do not make the decisions.

2. Constructability, which lives outside the model

Whether a bolt can be reached with a spanner once the adjacent member is up. Whether an assembly fits on a trailer. Whether the erector wants the splice at a different point because of crane reach. Whether the shop can actually roll that plate thickness. None of this is in the design file. It is in the heads of people who have been to a fabrication shop and a site, and it is the difference between a drawing that is correct and a drawing that is buildable.

3. Responsibility and sign-off

Someone has to be answerable for the approval and fabrication documents. Codes of practice are explicit about where responsibility sits between the engineer of record, the fabricator and the detailer, and about how delegated connection design is submitted and approved before fabrication (AISC 303). Indian work carries the same structure through IS 800 and the contract documents. A generated drawing does not accept liability, and a client will not accept "the software produced it" as a defence.

4. Catching the systematic error

This is the part that most detailers underestimate. Automation does not make errors randomly; it makes them consistently. A wrong parameter, a stale template, a numbering setting changed after work started, and the mistake appears identically on every drawing, every BOM line and every NC file. Tekla's documentation warns that contour and pop-mark settings must be fixed before numbering and drawing creation because both affect numbering. As generation gets faster, the cost of a bad input rises, because it propagates further before anyone looks at it.

Hugh Dobbie's account of checking, written decades ago, has aged remarkably well: checking is a quality control function carried out by someone who has advanced from detailing through experience and ability, and the detailer then back-checks the checker's comments and remains responsible for accuracy (AISC Engineering Journal). Replace "drawing" with "generated drawing set" and the sentence still holds.

5. Judging when the automated answer is wrong

An automatic connection returns a result on almost any geometry you give it. Knowing that a result is dimensionally valid but practically wrong, because of erection sequence, weld access, a mismatch with the shop's preferred details or a load path the tool did not see, is an engineering judgement. The tool does not tell you it is uncomfortable.

The shape of the job in five years

Fewer people producing drawings. More people specifying, auditing and taking responsibility for what gets produced.

That shifts value toward roles that already exist and are chronically short: checker, connection specialist, model auditor, standards owner, coordination lead. The PMI talent-gap reporting makes the same point from the other direction, listing digital engineering, sustainable construction, contract management and multidisciplinary collaboration as the capability gaps in Indian construction, and noting that graduates arrive without practical project exposure (Outlook India). Automation does not fill any of those gaps. It widens the gap between people who only operate software and people who understand the work.

What to learn in the next two years

In rough order of return on effort:

Connection behaviour, not just connection commands. Why a haunch is proportioned the way it is, how forces reach the bolts, what happens to a moment connection when a member size changes. This is what lets you overrule an automatic result.
Checking as a discipline. Dimension chains, hole counts against mating parts, mark logic for handed pieces, weld symbols, BOM reconciliation. Learn to check a generated set faster than you could have drawn it.
Fabrication and erection reality. Time in the shop and on site. What a fit-up problem looks like before the crane is standing idle.
Model data quality. Profiles, materials, finishes, assembly logic, numbering rules and NC output. Generation amplifies whatever data discipline you have.
Custom components and parametric setup. The people who build the libraries that generation runs on are the ones automation promotes rather than replaces.
Reading a project, not a drawing. Revision control, RFIs, client standards, who decides what, and how to write a question that gets a usable answer.

Notice what is not on the list: memorising more commands. Software fluency is the entry ticket. It stopped being the differentiator some time ago.

What this means if you are starting out

The entry-level path is narrowing, not closing. The way through it has changed: arriving with the software on your CV is no longer enough, because the software is increasingly doing that part itself. Arriving having modelled a complete building, produced its drawings and bill of materials, found your own errors and understood why the connections are what they are, is what makes a fresher useful in a team that now spends more of its time checking than drawing.

That is also the honest answer to the question this post started with. The job is not disappearing. The junior version of it is being compressed, and the senior version of it is getting bigger. The fastest route through the compressed part is to be able to think about steel, not only to operate a model.

Where STEEL fits

STEEL is a Tekla Authorized Training Centre. Our 12-week program is 70% practical and ends with a complete PEB building: model, connections, GA and fabrication drawings, bill of materials, and the QA habits that decide whether a set is releasable. If you are an experienced engineer planning your next step, tell us where you are now and we will tell you honestly whether we can help.