Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallAn AI agent is not automatically the right tool for a predictable workflow. When a task has structured inputs, exact answers, or a small set of explicit rules, ordinary code may be simpler to test and maintain. Dimitris Kyrkos makes that case in a September 28, 2026 DEV Community opinion article. Its headline’s claim that half of production agents are effectively if-statements is not backed by a survey or prevalence estimate, so treat it as a provocation, not a measured statistic.
What is the argument against using an agent by default?
Kyrkos’s point is not that language models are useless. It is that adding an LLM, vector search, or an agent loop to a workflow can make it more complex without helping when the task is already deterministic. The useful question is: is a model the simplest thing that does this reliably?
As an Amazon Associate I earn from qualifying purchases.
The article offers three illustrative cases—fixed-format extraction, exact data lookup, and rule-based support routing. They are engineering examples, not measured production studies. The article reports no comparative results for accuracy, latency, cost, or maintenance.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →When can a regular expression handle extraction?
If an invoice identifier always consists of INV- followed by eight digits, a regular expression can check that pattern directly. A model may be useful for invoices with varied layouts or ambiguous text, but it need not be the first step for a value with a known format.
Kyrkos uses a hypothetical 98% success example to illustrate a possible fallback design. It is not a benchmark or a reported error rate. In practice, validate extracted values against the expected format and define what happens when the check fails—such as sending the item for human review—rather than treating a plausible model response as verified.
When should exact lookup use SQL instead of vector search?
Consider a request for every unpaid order from one customer during a particular date range. That is a request for records matching explicit conditions. Kyrkos argues that structured database filters are a better fit for exact, complete, and auditable results than semantic retrieval.
Vector search is designed for similarity-oriented retrieval—for example, finding documents that discuss a concept even when they use different wording. It is not a substitute for an exact filter when the user needs all records satisfying defined fields. The distinction is about the query: exact facts call for explicit conditions; fuzzy meaning can call for semantic retrieval.
When are ordinary branches enough for support routing?
If a support workflow has a small, known set of conditions, those conditions can be expressed as ordinary branches. Kyrkos contrasts that approach with an agent loop that plans, invokes tools, and uses memory. When the possible decisions are already enumerable, explicit rules can make the outcome easier to explain and reproduce.
An agent loop becomes more relevant when the workflow must interpret varied requests, choose among changing actions, or handle ambiguity that is difficult to encode as fixed rules. The trade-off should be deliberate: flexibility can come with less predictable behavior and more components to manage.
How can you decide whether an AI component belongs?
- Check the input and output. Are inputs structured, and is the required output fixed-format? If so, consider validation and conventional parsing first.
- Identify the kind of question. Does it require exact facts from defined fields, or similarity and interpretation across unstructured material?
- Count the decisions. Can the workflow’s conditions and outcomes be listed in advance? If they can, compare a rules-based implementation with an agent.
- Plan for errors. Decide how a wrong result is detected, contained, corrected, and escalated before choosing an approach.
- Account for ownership. Ask who will maintain each added dependency and whether its added capability justifies that work.
- Compare implementations on evidence. Measure task accuracy, latency, cost, failure handling, reproducibility, and maintenance burden for the actual workflow. Kyrkos’s article supplies no measurements on these dimensions.
Does this mean you should avoid AI?
No. Kyrkos’s stated boundary is to use AI where ambiguity lives—such as unstructured text, fuzzy matching, or generation—and not to make it the default for work with deterministic rules. A practical design can also combine approaches: apply code to enforce known formats or exact constraints, and use a model where interpretation is genuinely needed.
The headline is best read as a challenge to examine architecture, not as evidence that a particular share of deployed agents are unnecessary. The article provides no survey, sample, method, or industry-wide estimate.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
