AI Agents vs Chatbots: Most Tasks Don't Need an Agent

An agent is a chatbot with a while loop and a credit card.

That is the whole difference. Everything else in the marketing is set dressing.

You ask a chatbot something, it answers, it stops. You give an agent a goal, and it plans, calls a tool, reads the result, notices it got something wrong, calls another tool, and keeps going until it decides it is done or until you kill it. The loop is the product. The loop is also the bill.

Gartner expects more than 40% of agentic AI projects to be cancelled by the end of 2027, blaming runaway cost, unclear value, and weak risk controls. In the same release they reckon only about 130 of the thousands of vendors selling "agentic AI" are doing anything actually agentic. They have a name for the rest: agent washing.

Three things get crammed into one word

They are not the same thing, and the confusion is where the money leaks.

Level 1, the chatbot. One call in, one answer out. It touches nothing.

Level 2, tool calling. You hand the model a list of functions. It picks one, your code runs it, you feed the result back. Your code still decides what happens next. Anthropic calls this a workflow: "systems where LLMs and tools are orchestrated through predefined code paths."

Level 3, the agent. Now the model decides what happens next. Same source: "systems where LLMs dynamically direct their own processes and tool usage, maintaining control over how they accomplish tasks." No fixed path. It loops.

Most of what gets sold to you as an agent is Level 2 with a nicer dashboard.

Now the part that never makes the sales deck. A May 2025 paper found that agents fail at a roughly constant rate per minute of human work. A constant failure rate means success falls off exponentially the longer the task runs. Every agent has a half life.

Do the arithmetic once and you never forget it. An agent that is right 95% of the time on a single step is right 0.95^20, about 36%, across twenty steps. That is not a flaw in the model. That is just multiplication.

What the benchmarks actually say

Read that table top to bottom and you will see a gradient hiding in it.

The closer a benchmark gets to a real job with real consequences, the worse agents do. Curated coding problems the model has probably already seen: 95%. The same kind of problems in repos it has not seen: 46%. Actual office work with finance and admin in it: 30%. And the number that matters most for anything you plan to leave unattended, the same task eight times in a row with zero misses, falls off a cliff.

The tau-bench figure is old and I am flagging it as old. But the 2026 reliability work reproduces the shape exactly: pass@1 flatters agents, repeated trials expose them.

The cost problem nobody prices in

Every step of the loop resends the entire conversation so far. Context grows with the number of steps, and you pay for the whole context on every step. So your input bill grows roughly with the square of the step count. Double the steps, quadruple the input tokens.

Anthropic's own engineers put a number on it: agents use about 4x the tokens of a chat, and several agents working together use about 15x.

At current Claude API rates, Sonnet 5 is $2 per million input tokens and $10 per million output. Sounds like nothing. Then it runs for six hours.

Real bills, all from TechCrunch, 5 June 2026:

And here is what people miss. The expensive failure is not the wrong answer. It is the agent that gets stuck in a loop, retries the same broken call two hundred times, and charges you full price for producing nothing. A chatbot that fails costs you a fraction of a cent. An agent that fails can run all night.

The Algerian reality

An agent needs three things before it can do anything useful: an API to call, structured data to read, and a way to check its own work.

Count how many of those the average Algerian merchant has.

Payment rails. GIE Monétique's 2025 report, published February 2026, counts 644 web merchants integrated into the national online payment platform. In the whole country. Meanwhile the CNRC counted 2,419,913 registered operators at 10 December 2025, of which 1,092,176 are retail merchants.

Six retail merchants in every ten thousand can take a card payment online. That is the digital economy your agent is supposed to be acting inside.

Data. Stock lives in a notebook or a WhatsApp thread. Orders come in as Instagram DMs. Returns get settled out loud with the delivery driver. There is no database to query, no order table, no state to reason over. An agent with nothing to read is a very expensive chatbot.

Language. On DialectalArabicMMLU, 19 Arabic capable open weight models averaged 51.9% on Modern Standard Arabic and 45.0% on Maghrebi. Those are small open models, not frontier ones, so do not over read it. But your customers write in Darija, in Latin letters, with French words dropped in, and none of that makes the loop more reliable.

And the bill is in dollars while your margin is in dinars.

One rail does exist here, and it is worth naming: delivery. Yalidine, ZR Express, Maystro and Noest all expose APIs, and Yalidine's is genuinely well documented. If any automation earns its keep in this market, it starts there. Not with an agent. With a script.

Where agents actually earn their keep

Three profiles, and they are narrow.

Coding with tests. The output is machine verifiable, retries are cheap, and the work is expensive to do by hand. This is the one place agents are unambiguously making money in 2026.

Deep research. Anything where a human would otherwise spend four hours reading sixty pages. It parallelizes, it tolerates noise, and you check the result at the end.

Support, but only on top of real documentation. Intercom reports Fin averaging 76% automatic resolution across 12,000 customers. That is a vendor number, so treat it like one. But read their own caveat: deployments that neglect the knowledge base stall at 30% to 45%. The agent's ceiling is set by the documentation you feed it. No documentation, no agent.

And the honest counterweight: Klarna replaced 700 agents' worth of support with AI, then started hiring humans back in 2025. Their CEO's words: "We focused too much on efficiency and cost. The result was lower quality, and that's not sustainable."

So when should you actually reach for one

Use an agent only when all four are true.

1. The task takes a human more than 30 minutes.
2. The output is verifiable, by a machine or by you in under a minute.
3. A wrong answer is cheap to throw away.
4. An API already exists. Not "we'll build one." One that is live today.

If even one of those is false, ask the model your question directly and get on with your day. That covers most of what you need.

If you do run one: hard cap on steps, hard token budget with a kill switch, log every tool call, and never let it write to your stock or trigger a shipment without a human pressing the button first.

Anthropic's own advice, from the company that sells you the models: "find the simplest solution possible, and only increase complexity when needed. This might mean not building agentic systems at all."

The bottom line

An agent is a chatbot with a while loop and a credit card. The loop is the value, and the loop is the bill. The best agents alive finish about 30% of real office tasks unsupervised, they burn roughly 4x the tokens of a chat, and cost climbs with the square of the loop length, so a stuck one charges you all night for nothing. Here in Algeria, 644 merchants out of 1.09 million can take an online card payment, which means the APIs and the clean data an agent runs on mostly do not exist yet. So most days: ask the question, take the answer, move on. The marketing pushes agents because agents are what sells.