AMAZON JUST KILLED MECHANICAL TURK. THE MANUAL SDR IS NEXT.

Amazon announced this week that it is closing Mechanical Turk to new customers. Twenty years ago, AMT was genuinely clever. The insight: there are tasks computers cannot do cheaply, but humans can. Label this image. Transcribe this clip. Categorise this data point. Pay them a few pence. Do it at scale.

The problem is that AI can now do most of those tasks faster, more cheaply, and more consistently than any crowd of humans. So the crowd is being stood down.

This is a small story in most industries. In B2B sales, it is a signal worth paying attention to.

The SDR was always the Mechanical Turk of go-to-market

Think about what a junior SDR actually does with most of their day.

Find a prospect in Sales Navigator. Copy their name, title, and company into a spreadsheet. Check whether the company is hiring. Open a template. Write a version of the same message sent yesterday. Click send. Log it in the CRM. Repeat.

That is not sales. It is data labelling. The cognitive load of each task is low. The value comes from volume and consistency. The human's advantage over a machine was only ever that the machine had not been trained to do it yet.

That training has happened. The machines are ready.

What our own pipeline shows

We run an AI outreach system on behalf of clients. The AI finds companies, drafts messages, and manages conversation threads on LinkedIn. A human approves every message before it sends. Nothing leaves the account without a sign-off.

The numbers from our live pipeline.

On our own LinkedIn account, the system has opened around 700 conversations since June and booked 20 meetings, 8 of them in the last 30 days. Along the way the orchestration layer has made hundreds of decisions: advance this lead, draft this follow-up, wait for a reply, close this thread. Every one a judgement a human did not have to make.

The conversion pair: 9% of people who receive the opening message reply. 30% of those who reply book a call.

One human operator. One hour a day on approvals. Not a team of SDRs.

The point is not that AI is impressive. The point is that the unit economics have already crossed the threshold. At the rates founders are paying for SDR salaries, commission, and management overhead, the case for running the repetitive layer on AI is not a future bet. It is a present calculation.

This is not about removing judgment from sales

The Mechanical Turk narrative gets used to make AI adoption sound like a cost-cutting exercise dressed up as innovation. That framing misses what is actually happening.

AMT shut down because the tasks it was built for no longer need humans at all. No judgment required. Labelling a cat photo is a closed task.

Outbound sales is not a closed task. A reply that does not fit the template requires a human. Timing a follow-up after a prospect posts about a new hire requires a human. Deciding whether to call or message after three LinkedIn touchpoints requires a human.

The AI handles the parts that do not require those things. The human handles the parts that do. The role moves up the value chain rather than disappearing.

That is the distinction that matters when founders ask whether this is right for their business. The question is not "will AI replace my sales team." It is "which parts of your sales team's day should a machine be doing."

Four moves worth making now

Audit your current outreach for task versus judgment. Go through a typical sales day and separate every activity into two buckets. Repeatable rule, predictable output: candidate for automation. Contextual read, no fixed answer: keep it human. Most outreach processes are 70% task and 30% judgment. The 70% is where to start.

Build the approval layer before you build the volume. The biggest failure mode in AI-assisted outreach is removing the human too early. Every system we run has a human sign-off on every message. This is not inefficiency. It is the quality gate that makes scale trustworthy. Wire the approval step first, then increase the volume.

Measure decision volume, not just output. The metric most founders track is meetings booked. The metric worth adding is how many qualified decisions the system made this month: leads assessed, threads advanced, messages staged, disqualifications logged. Ours made more than 700 last month. That is a truer picture of what the system is doing than any single conversion number.

Pick one stage to automate before the whole funnel. Do not try to automate prospecting, outreach, follow-up, and qualification simultaneously. Pick one stage, set consistent rules, review the output for two weeks. Get the data. Then extend. The compound effect builds from one stage, not from five half-finished automations.

The broader shift

The Mechanical Turk shutdown will not appear in any strategy document. It is a quiet administrative decision about a product that had been running on borrowed time.

But it is the cleanest possible signal of a transition that is already well underway. Repetitive human work, where the human's only advantage was that the machine was not yet trained to do it, is being phased out category by category.

In B2B sales, that transition is not theoretical. The tools exist. The pipelines are running. The conversion rates are measurable and improving.

The founders who move now will not just save on payroll. They will build a prospecting system that compounds. Every conversation informs the next. Every reply pattern becomes a signal. The system learns which messages work in which sectors at which funnel stage.

That is the thing Mechanical Turk never was. AMT was labour arbitrage. What we are building is intelligence that accumulates.

AI Sales is one of three systems we run for founder-led B2B businesses. See how it works.

Want to see the pipeline running on your own LinkedIn? Book 20 minutes here.