Toptal AI Review: Toptal Now Sells Finished Work. Who Answers for the Result?
Related service: Staff augmentation services
Toptal AI is the AI services company Toptal launched on 9 November 2026, and it changes what you buy from Toptal: finished AI work delivered by Toptal's own team, rather than access to a freelancer you then direct. The launch release counts more than 200 PhDs in AI-related fields, 100 forward-deployed engineers, 15 new product lines and signed agreements with 14 technology partners, Anthropic, OpenAI and Lovable among them.
I'm the CTO of Conectia, and Conectia competes for the same buyers: we place senior engineers through staff augmentation and build AI systems for clients, so a Toptal that sells finished work moves closer to what we sell. Read what follows with that in mind. Every figure below is linked to a document published between 13 October and 9 November 2026, and where the record favours Toptal I've left it in.
If you searched «what is Toptal AI» or «is Toptal still a freelance platform», the short answers: Toptal AI is a full-stack AI services company launched by Toptal that plans, builds, evaluates and operates AI systems for AI labs, enterprises and startups. And yes, Toptal is still a freelance platform: the same release describes Toptal as a network «connecting businesses with the top 3% of freelance talent on demand». One brand now sells two different contracts, and the question worth your time is which one you are signing.
Toptal AI sells outcomes, and says so in its own words
The release and the Toptal AI site describe five things:
- The claim: it develops, evaluates and deploys «frontier-grade AI» on open-weight and proprietary models, while the client keeps «full ownership and control» over its AI systems and data.
- The people: more than 200 PhDs in AI-related fields and 100 forward-deployed engineers.
- The products: 15 product lines, among them AI Workstation (a control centre across vendors), Maestro (builds and manages models and agents) and RL Refinery (reinforcement-learning environments).
- The partners: Anthropic, OpenAI, SpaceXSI, Gemini, Copilot, Amazon Nova, Databricks, Agentforce, ElevenLabs, Lovable, Miro, Fin, Glean and Writer. Toptal AI calls itself vendor-neutral.
- The operation: MLOps, LLMOps and managed services, which the site says cover service levels, incident response and drift monitoring.
The site puts the pitch in four words: «One Hand to Shake.»
Notice what the launch counts first: people. A company selling frontier AI leads with 200 PhDs and 100 engineers, because an agent inside somebody's workflow still needs a person who answers for it. That is the thesis this blog keeps defending, and the launch is one more piece of evidence for it.
A month earlier, on 13 October, Toptal launched ExpertCrowd, a platform for RLHF, labeling and model evaluation that, according to its release, had generated over $5 million in recurring revenue from frontier labs and companies such as Scale AI before it officially launched. Toptal now sells expertise twice: as training input to the labs, and as finished systems to everyone else.
Toptal is still a freelance platform; Toptal AI is a second contract on top
Nothing in the launch retires the marketplace. Toptal's boilerplate still describes a freelance network that has served more than 35,000 clients in 140+ countries, and its page for hiring forward-deployed engineers lists «Freelance Forward Deployed Engineer (FDE)» profiles, engaged by the hour or full-time after a trial of up to two weeks.
The same words now cover two purchases. In the first, you hire a person through Toptal and you direct the work, which is how a talent marketplace has always operated. In the second, you buy an outcome, an agent in production or a fine-tuned model, and Toptal AI's team delivers it. The first leaves the result on your side of the table. The second promises to move it to Toptal's. Which one you signed decides who answers when the agent misroutes a refund at two in the morning.
Upwork's revenue fell 8% and Fiverr lost one buyer in four: the open marketplace is shrinking
Toptal is private. Upwork and Fiverr publish quarterly, and both reported in the same weeks.
Upwork's third quarter, reported on 2 November: revenue of $185.6 million, down 8%; gross services volume of $914.1 million, down 10.1%; 717,000 active clients against 794,000 a year earlier, one client in ten gone. Its quarterly filing attributes the fall to difficulty winning and keeping clients, driven by «the evolving impact of AI on certain categories of freelance work», by search changes that hurt organic acquisition, and by macroeconomic uncertainty.
Fiverr's third quarter, reported on 29 October: revenue of $90.9 million, down 15.8%; 2.4 million annual active buyers against 3.3 million, down 25.8%, roughly one buyer in four. Fiverr expects «the shift from search to AI-driven discovery» to drive a steeper decline in marketplace traffic through 2027.
The numbers that cut the other way belong here too. Upwork's volume per active client rose 8% to $5,418, volume from freelancers with AI skills reached an estimated annualised run rate of nearly $450 million, up 55%, and the results met or beat analysts' expectations, according to Staffing Industry Analysts. Fiverr's spend per buyer rose 12.9%. Fewer buyers, each spending more, on harder work: the shape our August read of Upwork's second quarter found, and that post's caveat holds: these are Upwork's and Fiverr's figures, not Toptal's.
Upwork bought an employer and Toptal hired forward-deployed engineers: both moves add responsibility
With its results, Upwork announced that Lifted, its enterprise subsidiary, had acquired WorkMotion, a Berlin employer of record that «will directly employ contingent talent across more than 30 countries». The filing puts the price at about $129.7 million in cash, closed on 30 October.
Upwork adds an entity that employs people; Toptal adds a company that delivers results. Neither release explains a motive, and I won't supply one. What they share is on the page: the marketplace introduction is the line that is shrinking, and employment and delivery are the lines both are adding.
We've seen this movie. Turing made the same turn earlier, from remote-developer staffing toward AI work for labs and enterprises, as we reviewed in August. And the forward-deployed model never worked because of the title. It worked because an organisation stood behind the deployed engineer, with escalation, rotation and a planned exit, the doctrine I described in Staffing ends at the match, deployment ends at the handover. A forward-deployed engineer without that backing is a freelancer with a new title.
When the freelancer becomes a forward-deployed engineer, the contract has to name who answers
In a marketplace engagement, accountability sits with you: you chose the person and you direct the work. In an outcome engagement it moves, but only as far as the paper moves it. Five clauses decide where it lands:
- Employment. Whether the engineer is an employee or an independent contractor decides IP assignment, confidentiality and what happens on the day they leave. The launch release does not say how Toptal AI's 100 engineers are engaged.
- Acceptance. An outcome needs a definition of done: acceptance criteria, the evals the system must pass, and the name of whoever signs them off.
- Operation. Who holds the pager once the agent is live, under which service level and response time. Toptal AI lists service levels and incident response among its managed services; ask for the numbers.
- Ownership. You keep your systems and data, Toptal AI promises. Get it in writing for the fine-tuned weights, the eval sets and the training data, not only the code.
- Remedy. What happens when the outcome is not reached: rework at the vendor's cost, a credit, a right to terminate. A trial protects the first two weeks; it says nothing about month four.
An outcome you cannot accept, operate, own and remedy in writing is a staffing contract with a services label.
The case for Toptal AI is strong, which is why the questions matter
The strongest argument against this post is Toptal's scale. Fifteen years of screening data, more than 200 PhDs, 14 signed partnerships and one point of contact is what a buyer who does not want to assemble five vendors asks for, and a talent network moving toward outcomes is moving toward the accountability buyers say they want. The ownership promise settles one of the five clauses before you ask.
Nothing public says Toptal AI cannot settle the other four; the launch is five days old and its contracts are not public. What you can request is one named system in production, with its service level.
Six questions to ask Toptal AI, or any AI services firm, before you sign
- Who are the people on my system, by name, and who employs them?
- What is the acceptance test, and who signs it off?
- Who is on call once it is live, under which service level?
- What do I own on exit: code, weights, eval sets, data, runbooks?
- What happens when an engineer leaves mid-build?
- Do your technology-partner agreements shape what you recommend? A fair question for any integrator with partner agreements, and the contract is where «vendor-neutral» gets written down.
Ask us the same six. Our answers in one paragraph: the engineers are Conectia employees; we answer within 24 hours, and the first commit lands within 72 hours of the brief; within the first 30 days, a replacement is proposed within 7 days if the fit is wrong; and our AI development covers the whole cycle, from the build to inference, guardrails and operation in production.
Questions buyers ask about Toptal AI
What is Toptal AI?
An AI services company Toptal launched on 9 November 2026 to plan, build, evaluate and operate AI systems, with more than 200 PhDs, 100 forward-deployed engineers, 15 product lines and signed agreements with 14 partners including Anthropic, OpenAI and Lovable.
Is Toptal still a freelance platform?
Yes. Toptal's November 2026 release describes a network «connecting businesses with the top 3% of freelance talent on demand», and its hiring pages still list freelance profiles. Toptal AI is a separate offering that sells delivered outcomes.
Is Upwork's revenue declining?
Yes. Upwork's third-quarter 2026 revenue was $185.6 million, down 8% year over year, with gross services volume down 10.1% and active clients down from 794,000 to 717,000. Volume per client and AI-related work both grew.
What are the alternatives to Toptal for AI work?
Talent platforms with AI lines, AI consultancies, and engineering firms that employ their engineers and run the system in production. Our guide to the best Toptal alternatives in 2026 compares them by model.
One hand to shake is a good promise; read whose name is on the result
Toptal has moved from selling access to selling outcomes while the open marketplaces beside it lose buyers. That is good news for anyone buying AI work, provided the contract follows the press release. Before you shake the one hand, find out whose name is on the result in month four.
If you want to hear how we answer the six questions for your system, talk to a CTO.
Sources:
- Toptal via PR Newswire. Toptal AI Launches, Delivers Frontier-grade AI With Open-weight Models and Full Data Ownership (9 November 2026). prnewswire.com · toptal.com
- Toptal AI. Home page (read 14 November 2026). toptal.ai
- Toptal via PR Newswire. Toptal Launches ExpertCrowd to Train AI with Vetted Domain Experts, Releases Frontier SWE Benchmark (13 October 2026). prnewswire.com
- Toptal. Hire Forward Deployed Engineers (FDE) (read 14 November 2026). toptal.com
- Upwork. Third-quarter 2026 results, Exhibit 99.1 to Form 8-K (2 November 2026). sec.gov
- Upwork. Form 10-Q for the quarter ended 30 September 2026 (2 November 2026). sec.gov
- Fiverr. Fiverr Announces Third Quarter 2026 Results (29 October 2026). fiverr.com
- Staffing Industry Analysts. Upwork Q3 revenue falls 8.0%; subsidiary acquires EOR provider WorkMotion (3 November 2026). staffingindustry.com



