Onboarding: why the first 90 days matter so much for autistic employees

Autistic Mirror (outside view, role employer)

The first weeks are neurologically expensive because a new work environment means a flood of unresolved prediction errors for the autistic brain. Predictive Coding describes how the brain constantly compares its expectations with what actually arrives and evaluates the difference between the two. Autistic processing weights the actual sensory input more heavily than its own prediction. That means: where a neurotypical brain quickly bundles unfamiliar stimuli, routines and social rules into rough patterns and tunes them out, your new colleague processes every detail with full attention, because the prediction model for this environment does not yet exist. Unfamiliar office noise, implicit communication norms in the team, where things are filed, when to ask and when not, how your manager distinguishes praise from criticism: in the first weeks all of this is open computational workload.

The second system involved is Autistic Inertia, the biologically higher threshold for switching between states, because the basal ganglia and resting-state circuits are wired differently. Structurally, onboarding is nothing but a week of continuous state changes: new topic, new person, new software interface, new kind of task. Every switch costs disproportionately much. Alongside this, masking often runs in parallel, meaning the active effort to adjust one's own behaviour so that it does not stand out in the new social environment. Masking costs prefrontal resources that are then missing for processing the actual work content.

The result is that after an onboarding that ignores all of this, what gets measured is often not poor performance but a state of exception. Your colleague is then working with a much smaller neurological budget than in a stable environment. Concluding from this that the job is not a fit or that performance is too low is a fallacy: the onboarding created the conditions under which performance is difficult, the person did not reveal her limit.

If onboarding fails and the first weeks are set as the reference for future expectations, a further problem arises through the Double Empathy Problem. It describes that misunderstandings between autistic and non-autistic people are mutual, because both sides bring different communication styles. If your team learns during these weeks to read your colleague as "difficult", "distant" or "slow" because communication was bumpy, these interpretations become the foundation of the relationship. Correcting them later costs disproportionately more than creating a structured foundation from the start.

What makes the first weeks decisive: every stable element you introduce now permanently reduces the computational load. Written instructions, a predictable daily structure, clear contact persons for clear questions, and explicit communication instead of implicit expectations are not support measures for a weak person. They are conditions under which the prediction model for your environment is built faster and more stably, so that from week two, three and four your colleague arrives with a budget for her actual work instead of a budget for basic social and sensory orientation.

Autistic Mirror (outside view, role employer)

After an induction week with close support, the structure was not just pleasant, it actively provided orientation: who asks whom? What comes next? What is expected today? When this structure falls away, many autistic employees do not face a neutral situation, but one with considerable additional cognitive effort.

The central mechanism is Predictive Coding. The brain constantly compares what it expects with what actually arrives. Autistic processing weights the actual sensory and situational state more heavily than its own prediction. That means: as long as someone is there explaining the daily routine, there is a reliable external source of prediction. The support person took on this work. As soon as they are gone, the brain has to derive the predictions itself from incomplete knowledge, and small deviations between expectation and reality keep arising. Each of these deviations costs processing energy.

This has a concrete effect on starting tasks. This is where Autistic Inertia comes in, the neurologically based threshold for switching between states. Starting, stopping, switching runs via differently wired regulatory circuits, and it is neurologically expensive when the next step is not clearly set from outside. During the induction week, the support person often provided this impulse implicitly: by arriving together, by "we are doing X now", by their mere presence as a rhythm. When that falls away, the employee has to initiate every start herself, and that is exactly the neurologically demanding part.

Intolerance of uncertainty amplifies this as a precondition: the feeling of not knowing whether one is doing the right task, whether the prioritisation is right, whether one is doing something wrong without noticing, whether one may ask and if so whom and how: these are all open loops. An open loop is a persistent prediction error: the brain keeps an expectation active that has no resolution yet, and the comparison keeps running and permanently uses capacity, even without consciously thinking about it. During onboarding, the support person closed these loops almost automatically. Now nobody closes them anymore.

What you might see from the outside does not have to look like overload. Many autistic employees mask and compensate through increased effort. But this is a short-term solution with costs: less capacity remains for the actual work, and over time this builds up into exhaustion.

What helps structurally are not big interventions. Clear written documentation of tasks and their priorities that the employee can read at any time without having to ask. A reliable contact person with defined times instead of open availability, so that the question "may I disturb now" does not cost every contact. And a short weekly check-in with a concrete daily structure considerably reduces the load of uncertainty without close support being needed permanently. The structure does not have to come from a person, it can be transferred into processes and documentation.